Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Contents lists available at ScienceDirect

Journal of Behavioral and Experimental Finance

journal homepage: www.elsevier.com/locate/jbef

Full length article

Technical market indicators: An overview

Jiali Fang a,∗ , Yafeng Qin a , Ben Jacobsen b
a
School of Economics and Finance, Massey University, New Zealand
b
University of Edinburgh Business School, United Kingdom

article info abstract

Article history: Current evidence on the predictability of technical analysis largely concentrates on price-
Received 2 September 2014 based technical indicators such as moving averages rules and trading range breakout
Accepted 18 September 2014 rules. In contrast, the predictability of widely used technical market indicators such as
Available online 4 November 2014
advance/decline lines, volatility indices, and short-term trading indices has drawn limited
attention. Although some market indicators have also become popular sentiment proxies
JEL classification:
in the behavioral finance field to predict returns, the results generally rely on using just
G10
G12
one or a few indicators at a time. This approach raises the risk of data snooping, since so
many proxies are proposed. We review and examine the profitability of a wide range of 93
Keywords: market indicators. We give these technical market indicators the benefit of the doubt, but
Technical analysis even then we find little evidence that they predict stock market returns. This conclusion
Market indicators continuously holds even if we allow predictability to be state dependent on business cycles
Sentiment indicators
or sentiment regimes.
Asset pricing
© 2014 Elsevier B.V. All rights reserved.

1. Introduction modern studies on technical analysis produce supportive

evidence of its profitability.
Technical analysis, a methodology for forecasting the However, some academics consider that much of the
direction of security prices through the study of past mar- positive evidence is pseudoscience or dubious (Paulos,
ket data, is widely used by practitioners. A survey on 692 2003) because of problems such as data snooping and
fund managers shows that 87% of the fund managers place sample bias. More importantly, most studies only consider
some importance on technical analysis when making their price-based technical indicators that are just a subset
investment decisions (Menkhoff, 2010). However, there of all technical indicators. There are other types of
has been controversy over whether technical analysis ac- technical indicators – so called technical market indicators
tually helps to predict the markets. Some previous studies – that investors and media and finance professionals use
(e.g., Brock et al., 1992; Fama and Blume, 1966; Jensen and frequently as well, such as advance/decline lines, the
Benington, 1970) try to examine the predictability of clas- Arms Index, and volatility indices. Contrary to price-based
sic price-based technical indicators. However, these stud- indicators that use only historical prices information to
predict future price movements of individual stocks or the
ies cannot provide affirmative evidence of the usefulness
aggregate market, market indicators use a variety of other
of technical analysis because the conclusions are mixed. In
financial market information, such as trading volumes,
a review study, Park and Irwin, 2007 find that 56 out of 95
investor sentiment survey results, and implied market
volatility, to predict the aggregate market.
∗ Correspondence to: School of Economics and Finance, Massey Market indicators are very important to practitioners.
University, Private Bag102904, North Shore, Auckland 0745, New Zealand.
A popular technical analysis book, Achelis (2001, p. 38),
Tel.: +64 9 441 8176x9242. states, ‘‘Market indicators add significant depth to tech-
E-mail addresses: j.fang@massey.ac.nz (J. Fang), y.qin@massey.ac.nz nical analysis, because they contain much more informa-
(Y. Qin), ben.jacobsen@ed.ac.uk (B. Jacobsen). tion than price and volume. . . the analogy being ‘all boats
http://dx.doi.org/10.1016/j.jbef.2014.09.001
2214-6350/© 2014 Elsevier B.V. All rights reserved.
26 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

rise in a rising tide’’’. Many major data vendors, such as across different market indicators. Hence, using the widest
MarketWatch and Bloomberg, report these indicators reg- range of market indicators we can obtain also safeguards
ularly as key market statistics. For example, MarketWatch, our results against data-snooping issues.
among many other data vendors, presents analysis on the Our preliminary analysis shows that at a conservative
Chicago Board Options Exchange (CBOE) volatility index 10% significance level, 30 out of 93 indicators show pos-
nearly every day. In contrast to the popularity and accep- sible predictability. However, only 10 of these remain sig-
tance these technical market indicators have received from nificant after we conduct sub-sample robustness checks.2
practitioners, however, limited academic scrutiny has been Since some indicators exhibit sign-switching predictability
presented on the predictability of these market indicators. in the sub-sample test, we employ rolling window regres-
Our study tries to fill the gap in the literature by examining sions to take a closer look at the stability of indication. This
the predictability of a wide range of market indicators. We further reduces the number of possible predictive indica-
look at all 93 market indicators from the Global Financial tors to eight. We then conduct economic significance tests
Data database,1 which, to our best knowledge, is one of the to account for risk and transaction costs and none of the
most comprehensive ranges of market indicators. technical trading strategies beats the naïve buy and hold
Branch (1976), which comes closest to our work, re- strategy in terms of either the Sharpe ratio or Jensen’s α .
views the predictability of a range of 10 market indicators We make no conclusions on the predictability of
and documents mixed results. The author points out that market indicators yet. A recent strand in the literature
the study is limited by insufficient data access, even many documents that some return-predicting models are time
more indicators are proposed. Nearly 40 years later, with varying and state dependent (Dangl and Halling, 2012;
access to a much broader set of market indicators, includ- Henkel et al., 2011; Jacobsen et al., 2007; Yu and Yuan,
ing those introduced during those 40 years, we can now 2011). For example, Jacobsen et al. (2007) show that an
conduct a comprehensive study on market indicators. Us- increasing industrial metal index can significantly predict
ing a wide range of 93 market indicators, however, we find higher stock market returns during contractions but lower
no evidence that they show predictability for future stock market returns during expansions. These two effects off-
returns. This conclusion consistently holds even if we allow set each other. As a result, the industrial metal index does
predictability to be state dependent on business cycles or not show any predictability if not contingent on states of
sentiment regimes. the economy. The underlying reason for the time varia-
Our range of indicators covers nearly all those available tion of return predictability is that the same news may be
in the Global Financial Data database except a few that interpreted differently by investors across business cycles
do not have sufficient data. In addition, we add common (Boyd et al., 2005; Jacobsen et al., 2007; McQueen and Ro-
transformations of the raw indicators provided by Global ley, 1993).
Financial Data. The idea is that the raw data may contain In the case of technical analysis, the same information
both signal and noise and the technical transformations can also be interpreted differently across different states of
might reduce the noise in those series. However, we find the economy. For instance, a rising value of investors’ bear-
no significant predictability either from the raw data or ish sentiment index during contractions often indicates a
from the popular technical transformations. Moreover, we bearish sentiment extreme that signals a potential mar-
use the longest sample for each indicator available from ket reversal. In contrast, a rising value of bearish sentiment
the database to best avoid the data-snooping problems during market expansions can signal investors’ fear about
pointed out by many previous studies (e.g., Lakonishok and the future market and thus a decreased market. If that is
Smidt, 1988). The longest sample in our study is nearly 200 the case, the observed non-predictability of the market
years and the overall average sample length is 54 years. indicators could be due to their time variation or state
Accounting for the data-snooping bias is particularly dependency. This conjecture has been recognized by prac-
important in the field of technical analysis. In particular, titioners3 and supported by the empirical findings on
the predictability of technical analysis can change over price-based indicators of Han et al. (2013), who show that
time. Brock et al. (1992) find that price-based technical moving average strategies generate much higher abnor-
trading strategies beat the market in their 90-year sample mal returns in recessions. We hence explicitly test if there
from 1897 to 1986. Fang et al. (2013), however, find is any time variation or state dependency that could also
that the predictability disappears completely on a fresh shadow the real predictability of market indicators by us-
25-year sample from 1987 to 2012. Hence, using the ing the regime-switching methodology of Jacobsen et al.
longest sample available best prevents such data-driven (2007).
results. On the other hand, predictability also varies greatly We define our business cycles by using National Bureau
of Economic Research (NBER) data. Our results show that

1 See www.globalfinancialdata.com. We exclude a few indicators from

2 These 10 market indicators are the NYSE short sales volumes—
the Global Financial Database, including New York Stock Exchange (NYSE)
money borrowed and NYSE debit balances, since the data end in 1967 members/specialist/total, the NYSE short interest ratio, the NYSE
and 1970, respectively. We exclude most CBOE volatility indices except advances/declines/new highs, the Alternext declines/new highs, and
that for the Standard and Poor’s (S&P) 500, since their sample periods are weekly NYSE cumulative highs.
shorter than 10 years, most not actually starting until 2011. For similar 3 Edwards et al. (2012, p. 17) point out, ‘‘One of the keys in long-term
reasons, we do not include the S&P total dividend declarations, the S&P chart analysis is realizing that market behaves differently in different
monthly dividend declarations, the S&P 500 monthly advancing/declining economic cycle. . . . Identifying where you are in an economic cycle. . . is
stocks, or the S&P 500 index sales per share. critical to interpreting the chart patterns evolving at that time’’.
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 27

26 (21) market indicators show possible predictability in We follow the same steps above as those under ordinary
expansion (contraction) periods in the first place and the least squares (OLS) and find consistent results of no pre-
numbers are both smaller than the 30 we find in the full dictability. We also check if outliers could affect our re-
sample. Further F -test results testing the statistical differ- sults by using robust regressions, particularly because we
ence in predictability between expansion and contraction observe problems such as widening confidence bounds
periods reveal that only 19 indicators show significantly under the OLS rolling window regressions. Only one indi-
different predictability in one of the business states over cator may possess predictive value when the effect of ex-
the others. That is, most indicators (74 out of 93) do not treme observations is controlled for. Moreover, we use the
seem to suffer from the problem where they potentially Chicago Fed National Activity Index (CFNAI) data alterna-
exhibit predictability in one of the business states but this tively to define business cycles; we also check the possible
predictability is shadowed by insignificant predictability impact of the recent 2008 financial crisis by omitting this
in the other business state in the full-sample test. Nev- period from our sample; lastly, we also replicate our analy-
ertheless, we continue our economic significance test for sis excluding the top and bottom 5% extreme observations.
those possible 26 (21) predictors under expansions (con- Our finding of weak predictability remains similar.
tractions) separately. During expansion periods, we find Can we rely on technical analysis? Our results, from
one market indicator has a higher Sharpe ratio than the evaluating a comprehensive range of technical market
buy and hold strategy and another market indicator has indicators using the longest sample available, do not seem
a significant positive Jensen’s α .4 During contraction pe- to suggest an affirmative answer to this question. Our main
riods, none of the 93 indicators show any economic value. contribution lays in filling the gap in the literature on a
Although it remains possible the two indicators have some comprehensive study of the technical market indicators
predictive value in expansion periods only, our overall re- by providing strong statistical and economic evidence on
sults generally do not seem to suggest the business cycle- their practical usage. Our study has particularly important
dependent predictability of the market indicators. implications for practitioners who rely heavily on technical
Besides testing the possible time-varying predictability analysis in making investment decisions.
on business cycles that many return predictability studies The rest of the paper is organized as follows. To best
have considered, we also look at the sentiment cycles re- serve the overview role of this paper, Section 2 first explic-
cently introduced by Yu and Yuan (2011). They find that itly reviews current evidence on our market indicators be-
the mean–variance tradeoff differs across high- and low- fore a formal analysis, including those from the sentiment
sentiment periods. The intuition for the differences be- field. This provides an overall understanding and expecta-
tween high- and low-sentiment periods on returns is as tion of the predictability of the market indicators. We then
follows. During a high-sentiment period, when more irra- introduce our data and methodology in Section 3, followed
tional investors participate, the price deviates more from by a presentation of the empirical results in Section 4.
its fundamental value, whereas during a low-sentiment Section 5 explicitly examines the time-varying predictabil-
period, the price more accurately reflects its fundamen- ity of the market indicators and Section 6 provides various
tals, with less sentimental noise. We thus are motivated robustness checks. We conclude the paper in Section 7.
to test if technical market indicators’ predictability can
differ across sentiment cycles, considering a fundamental 2. Technical market indicators
belief of technical analysis that the price has already re-
flected all information and investors’ aggregate sentiment Fundamentally, technical analysis believes that stock
is the main driver of price deviation from its fundamental prices follow trends because investors collectively repeat
value (Kirkpatrick and Dahlquist, 2010). The test results on their patterned trading behavior, which is the major driver
sentiment cycles remain largely similar to those on busi- of stock price fluctuations (Murphy, 1999). Although the
ness cycles: 21 (25) indicators show possible predictability patterned behavior may be irrational, by exploring the
in high-sentiment (low-sentiment) periods, while only 10 pattern – the trend – one can effectively anticipate future
of them show significantly different predictability across price movements.
these two regimes. Moreover, after we consider economic Market indicators can be classified into two groups:
significance, none of the market indicators remain predic- market sentiment indicators and market strength indi-
tive. These findings further eliminate the possibility of the cators. The market sentiment indicators predict market
state-dependent predictability of the market indicators. movements based on tracking the bullish or bearish psy-
We try to give our market indicators the benefit of the chology of the market. When bullish (bearish) sentiment
doubt as much as possible. In that sense, our linear regres- dominates the market, stock prices will rise (decline),
sion tests with a general correction for heteroscedasticity associated with an increasing demand for (supply of) se-
may be too restrictive. Therefore, as a robustness check, curities. Market strength indicators measure the strength
we verify whether these technical market indicators might – the breadth – of market movements. A strong movement
work if we reduce noise in the data. We model the het- with a high breadth reading will last longer and take the
eroscedasticity more explicitly using a GARCH(1, 1) model. market to higher highs or lower lows. Market indicators
expand the information set of technicians beyond classic
price and volume data to a variety of financial information.
4 The indicator US mutual fund equity fund redemptions has a higher Although market indicators are sometimes used in other
Sharpe ratio than the buy and hold strategy and the NYSE new highs markets, such as the futures market, they primarily analyze
indicator has a significantly positive Jensen’s α . aggregate stock market movements (Achelis, 2001, p. 31).
28 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

While evaluating overall market conditions remains im- Following this definition, we have 50 sentiment indi-
portant even when investing in individual stocks, the mea- cators in total and we further categorize them into 11 sub-
surements of individual stocks’ sentiment or the strength groups based on the type of raw information they use. Sen-
of price movements are generally noisy and less reliable timent indicators can incorporate various information that
with limited data access. past prices cannot reveal. Some use direct sentiment poll
Among the 93 market indicators our paper studies, 65 results, such as the American Association of Individual In-
are raw indicators that extract information from market vestors (AAII) and Investors Intelligence (II) sentiment in-
data directly, such as total market advance/decline issues dices, while others use data from the underlying deriva-
in a trading day, and the other 28 are transformed in- tive markets but not the stock market directly. For instance,
dicators that manipulate raw information through some we use put or call option volumes and volatility indices. In
formula. For example, net advances equal raw decline is- addition, a number of sentiment indicators use the statis-
sues deducted from raw advance issues. This has a ben- tics of different trading activities, such as odd-lot trading
efit for the practitioner. One may favor a few particular statistics, short sales statistics, mutual fund statistics, and
margin account balances. Lastly, the rest of indicators in-
transformed indicators when one believes that the trans-
clude Barron’s confidence index, American Stock Exchange
formation can provide further indications over what the
(AMEX) seat prices, and Moody’s or S&P 500’s aggregate
raw information can. For instance, in the above example,
number of positive/negative dividend news.
net advances indicate the strength of a trend on a relative
Technical theories impose different signs on the way
basis, comparing the up and down trend strengths, while
that the indicators predict the market, since they view dif-
the raw advances/declines focus solely on the up/down
ferent investors’ sentiment differently. A key issue is to
trend. While many previous studies use transformed indi-
distinguish between two major groups of market player
cators only – for example, on advance/decline information, sentiment: the uninformed and informed traders. Unin-
Brown and Cliff (2004) use the advance/decline ratio and formed traders are passive and trade for liquidity (Wang,
Zakon and Pennypacker (1968) use the advance/decline 2002). De Long et al. (1989, 1990) argue that uninformed
line – our study may shed light on whether using such a traders tend to act strategically on noisy signals and there-
transformation has an advantage over using the raw infor- fore their trading can affect prices in a systematic way that
mation or whether it actually masks the raw information’s deviates asset prices from fundamental values. According
true predictability. to Kirkpatrick and Dahlquist (2010), this group of investors
Table 1 summarizes our indicators. Panels A and B re- largely consists of individual traders. They lack sufficient
port our reviews of market sentiment indicators and mar- financial knowledge when making their investment deci-
ket strength indicators, respectively. We further classify sions and their behavior is that of a crowd; in other words,
our indicators into 14 sub-groups based on the type of in- they make decisions in line with everybody else, regardless
formation they use. As mentioned earlier, many indica- of the true financial facts. Thus, the theory refers them as
tors can use the same raw information. For example, our uninformed investors and views their decisions as always
first group of sentiment indicators uses option volumes wrong, such that trading against them will result in sig-
to proxy for aggregate sentiment, with a rising call (put) nificant gains. So-called contrarian indicators measure this
volume indicating investors’ bullish (bearish) sentiment group of investors’ sentiments. Examples of our contrarian
since they are hedging against a potential market rise (fall). indicators include odd-lot trading statistics, short sales of
Global Financial Data provides the information from two general public investors, and survey sentiment indices. In
sources, the CBOE and the OEX. The last indicator is a trans- contrast, informed traders, who are sophisticated investors
formed indicator that uses the ratio of the traded value of that have adequate financial knowledge, build their deci-
put to call options to measure the relative strength of bear- sions on precise analysis of the market. Theory views their
ish sentiment to bullish sentiment. This gives us a total of decisions as always true and accurate. Professional specu-
five indicators for this group. In the remainder of the paper, lators, position traders, hedge fund managers, professional
we discuss our results by such a grouping. To save space, arbitrageurs, and insiders are considered to be in this cate-
we review the underlying theory and existing evidence of gory (Kirkpatrick and Dahlquist, 2010). Typical sentiment
our indicators explicitly in Table 1 and give a brief overview indicators for this group of investors include volatility in-
dices, option trading volumes, and specialists’ short sales.
in the following.
These sentiment indicators have been receiving grow-
ing attention. In the academic fields, especially behav-
2.1. Market sentiment indicators ioral finance, many sentiment indicators used overlap with
those we examine, those used by technical analysis. Since
Sentiment is defined by Kirkpatrick and Dahlquist we have many indicators, we review the previous evidence
(2010, p. 90), as by group in Table 1. In brief, the previous evidence is mixed,
even when using the same sentiment proxies. For example,
The net amount of any group of market players’ op- Seneca (1967) finds that the short interest ratio predicts
timism or pessimism reflected in any asset or market monthly S&P 500 returns negatively from 1946 to 1965.
price at a particular time. When a stock or commodity Brown and Cliff (2004), however, suggest that the short in-
is trading at a price considerably above or below its in- terest ratio does not predict S&P 500 returns from 1965 to
trinsic value, something we will not know until consid- 1998. This raises the concern of data snooping and calls for
erably later, the difference or deviation from that value using a long sample period to safeguard against the data-
often will be accounted for by sentiment. snooping issue and to update the results. Moreover, many
Table 1
Overview of technical market indicators.
Market indicators Sample period Notes and formulas More references

Panel A: Market Sentiment Indicators

1. Option Volumes:
CBOE Calls Volume 1989–2011 This group of indicators uses data from closely-related option market instead of from stock market itself to Dennis and Mayhew (2002);
CBOE Puts Volume 1989–2011 predict returns. Due to financial institutions’ larger presence in the derivatives market, these indicators are Feldman (2010); Kurov (2010);
OEX Calls Volume 1989–2011 viewed as primarily a measure of institutional sentiment (Brown and Cliff, 2004). A call (put) option gives its Simon and Wiggins (2001).
OEX Puts Volume 1989–2011 holder the right, but not the obligation to buy (sell) the underlying security at a pre-specified price.
CBOE Ratio of Traded Value 1986–2011 Therefore, a rising call (put) volume indicates investors’ fear about a potential price increase (decrease), and
of Puts to Callsa reflects their bullish (bearish) sentiment. We study volumes of put/call options listed on Chicago Board of
Exchange (CBOE) and Standard and Poor’s 100 index (OEX), the volumes are published daily in their dollar
values. Different from using the volumes directly, the ratio of traded value of puts to calls is calculated from
dividing the CBOE daily dollar puts volume to the corresponding calls volume. It compares the relative
strength of bullish/bearish sentiment. The ratio is lower (higher) during bullish (bearish) sentiment periods.
While previous studies mainly focus on the put/call ratio only, Brown and Cliff (2004) and Wang et al. (2006)
both document that the put/call ratio does not predict stock returns.
2. Odd-Lots Volumes:
NYSE Odd Lot Purchases 1970–2011 The odd-lotters refer to small individual investors who usually trade at an amount less than the standard Abraham and Ikenberry (1994);
NYSE Odd Lot Sales 1970–2011 unit of trading. Their trading activities became popular contrarian indicators in the 1960s and the 1970s. Branch (1976); Brown and Cliff
NYSE Odd Lot Shorts 1970–2011 When they turn bullish (bearish), odd lot purchases (sales) increase in contrast to actual market declines (2005); Dyl and Maberly
(increases). NYSE publishes its odd lot trading statistics daily since 1970. Studies including Lakonishok and (1992); Gup (1973); Kaish
Maberly (1990), Zweig (1973), and Gup (1973) confirm the accuracy of odd-lot trading indicators. In (1969); Kewley and Stevenson
contrast, Brown and Cliff (2004) find that the odd-lot sales to purchases ratio does not predict market (1967).
returns, Neal and Wheatley (1998) also find that the ratio does not predict the size premium while the other
two sentiment proxies they use do.
3. Short Sales Volumes:
NYSE Short Sales-Members 1940–2008 Investors short sell to hedge against a future market decline. Technical theory insists that informed Glushkov (2006); Lamont and
NYSE Short Sales-General 1940–2008 investors (for example, NYSE members and specialists) always build their short selling decisions on reliable Stein (2003).
Public 1940–2008 savvy analysis, such that their short selling activities indicate future market declines. However the exact
NYSE Short Sales-Specialists 1940–2008 opposite holds for uninformed investors (general public), the market moves upwards when they short sale.
NYSE Short Sales-Total We study the weekly short sales information published by the NYSE. After 2008, The NYSE is no longer
compiling this data on a weekly basis. Instead, they report daily data of non-specialist, specialist and total
short sales. Since the definitions remain similar, and the weekly data before 2008 gives us a much longer
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

sample period, we focus on data before 2008. Reilly and Whitford (1982) find the specialist short sales ratio
(specialists’ short sales/total short sales) has no merit in predicting stock returns from 1971 to 1979, while
Fosback (1993) suggests the contrary that the ratio predicts the market during 1941–1975. However, Bowlin
and Rozeff (1987) argue the above two studies can be misleading since they both use nonindependent
observations, they correct this problem and find that the specialist short-interest ratio is inversely related to
subsequent NYSE stock returns. Nevertheless, Brown and Cliff (2004) and Branch (1976) both study the
specialists shorts and conclude they fail in predicting the stock returns. Branch (1976) also studies the
members’ short sales and finds no merit of it in predicting stock returns. Zweig (1973) on the other hand,
finds that floor traders (NYSE members) are subject to the same overemotional pressures as individual
investors so that they sell (buy) at bottoms (peaks) too. Lastly, the public short sales have drawn much less
attention in the literature, although practical technical analysis textbooks like Summa (2004) and
Kirkpatrick and Dahlquist (2010) view it as a contrarian indicator.

(continued on next page)

29
 30

Table 1 (continued)
Market indicators Sample period Notes and formulas More references

4. Short Interests:
NYSE Short Interest Ratioa 1931–2010 NYSE Short Interest Ratio = NYSE Short Interest Shares/Average Daily Trading Volume Biggs (1966); Gup (1973);
NYSE Short Interest Shares 1931–2010 NYSE Short Interest Shares represent the monthly number of shares investors have sold short, but have not Hanna (1976); Lamont and
yet covered or closed out for securities listed on the NYSE, NYSE Arca and NYSE Amex. Short interest shares Stein (2003); Mayor (1968);
predict the market by watching whether the majority of investors think the market is likely to fall. When the Shiller (2000); Smith (1968).
amount of short interest shares decreases (increases) that indicates the uninformed investors’ bullish
(bearish) sentiment, the market moves downward (upward). The NYSE Short Interest Ratio is a monthly
technical indicator calculated by dividing the NYSE short interest shares over the average daily trading
volume over the past 30 days. This ratio indicates how quick it takes to cover the short position in days. A
high ratio-that is, heavy short sell of uninformed investors—predicts a bullish market. Seneca (1967) and
Kerrigan (1974) find a significant negative relation between the short-interest ratio and future S&P 500
returns. However, Woolridge and Dickinson (1994) suggest a positive but insignificant relationship between
short interest ratio changes and individual stock prices. Similarly, Brown and Cliff (2004) and Vu and Caster
(1987) also find results that do not support the predictive role of the short interest ratio.
5. AAII/II Sentiment Indices:
AAII Bearish Index 1989–2010 Sentiment survey results provide a direct measure of investors’ view on the future market. American Han (2008); Sanders et al.
AAII Bullish Index 1989–2010 Association of Individual Investors (AAII) selects a random sample of its members each week to conduct a (1997); Wang et al. (2006)
AAII Neutral Index 1989–2010 sentiment survey, and the association gathers each participant’s opinion of the market movements in the
Investors Intelligence 1987–2010 next six months; up, down, or the same. The percentages of these opinions over the total responses form the
Bearish Percentage bearish, bullish and neutral sentiment indices. One vote per member is accepted for each weekly survey.
Investors Intelligence Bullish 1987–2010 AAII sentiment indices predict the market contrarily, as they represent the uninformed individual investors’
Percentage views. Similar to AAII indices, Investors Intelligence indices represent survey results run over more than 130
investment newsletter writers asking for their predictions about the market; the survey answers are
classified as bearish, bullish, or neutral. They are also contrarian indicators. Solt and Statman (1988) find no
statistically significant relation between the sentiment of investment newsletter writers and subsequent
stock returns. Using the same set of data, Clarke and Statman (1998) confirm the results, they also find past
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

returns and market volatility affect sentiment. In contrast, Lee et al. (2002) find the magnitude of bullish
(bearish) changes in sentiment leads to downward (upward) revisions on volatility and higher (lower) future
excess returns by using II indices as sentiment proxies. Fisher and Statman (2000) study both the AAII and II
indices. They find investor sentiment differ across groups, AAII indices predict the stock returns contrarily
while II indices play no predictive role. De Bondt (1993) also finds AAII indices predict future returns.
Table 1 (continued)
Market indicators Sample period Notes and formulas More references

6. Confidence Index
Barron’s Confidence Index 1932–2010 Barron’s Confidence Index = Malek (2000); Tabell and Tabell
average yield on 10 Barron’s top grade corporate bonds
Barron’s-a leading American financial magazine-uses
average yield on 10 Barron’s intermediate grade corporate bonds
(1964).
information from the bond market to calculate a sentiment index each week, and the index is referred to as
the Barron’s Confidence Index. As one of the most historically extensive sentiment indicators, it dates back
to 1932. Barron’s confidence index measures the discrepancy between the yields of high and low risk bonds
that are largely held by institutional investors (Gaumnitz and Salabar, 1969). These informed investors are
willing to accept a lower premium in yield for high risk bonds when they have more confidence with the
economy, indicating a rising index value. And the exact opposite holds as well. Ring (1974) supports the
view that the smart money will move from speculative to quality bonds during bearish market, and back
during bullish market, and he documents that the index is a better forecaster of tops than bottoms. In
contrast, Branch (1976) does not support Barron’s Confidence Index as a stable market predictor in different
sample frequencies, neither do Gaumnitz and Salabar (1969).
7. Exchange Seat Prices:
AMEX Seat Prices 1921–1993 Exchange seat prices capture professional investors’ sentiment. It rises when market activities and stock Chiang et al. (1987); You and
NYSE Annual Seat Price 1820–2003 prices rise, reflecting the prosperity of the brokerage industry, which provides professional services to the Holder (2008).
capital markets (Schwert, 1977). The AMEX seat prices are available monthly and the NYSE seat prices are
available annually. They focus on long term forecasting, while most other sentiment indicators capture
relatively short term variation. NYSE seat prices is the only annual sentiment indicator we examined, and it
also has the longest sample period across all of our sentiment indicators—it starts in 1820 which means
nearly 200 years of data. Keim and Madhavan (2000) document supportive evidence to the view that NYSE
seat prices capture market sentiment, although Schwert (1977) finds no evidence that the seat returns
predict market returns.
8. Volatility Indices:
CBOE S&P 500 Volatility 1986–2011 Volatility index, sometimes called the ‘‘fear index’’, measures the implied volatility for a group of near-term Feldman (2010); Fleming et al.
Index 2001–2011 put and call options related to a specific market index; that is, it represents investors’ expected risk of the (1996); Low (2004); Simon and
CBOE NASDAQ Volatility 1986–2011 market over the next 30 days. An increase (decrease) in the volatility index represents bearish (bullish) Wiggins (2001).
Index 2001–2011 sentiment of the market. For example, the S&P 500 volatility index hits its historic high of 89.53 on October
CBOE S&P 100 Volatility 24, 2008, reflecting investors’ serious concern during the financial crisis period. We view the volatility
Index 2005–2011 indices as primarily a measure of institutional sentiment because of institutions’ large presence in the
AMEX NYSE Arca NASDAQ derivative markets Brown and Cliff (2004). We study 5 different volatility indices that track investors’
100 Volatility Index sentiment for different stock indices. These include 4 CBOE published volatility indices for the S&P 500, the
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

CBOE DJIA Volatility Index NASDAQ, the DJIA and the S&P 100, respectively, and the last volatility index is published by the AMEX to
track the sentiment on the NASDAQ 100. Previous literature largely focuses on the S&P 500 volatility index
only. Whaley (1993) describes the construction of the S&P 500 volatility index in detail. The volatility index
firstly measures the implied volatility using prices of only 8 S&P 100 at-the-money put and call options, and
later the input expanded to options based on the broader S&P 500 index (Investopedia). Whaley (2000)
documents the S&P 100 volatility index as a reliable ‘‘investors fear’’ gauge, higher levels of the index
coincide with high degrees of market turmoil. Giot (2003) also shows a negative and statistically significant
relationship between the returns of stock and implied volatility indices on the S&P 100 and NASDAQ 100
indices. However Brown and Cliff (2004) document results that do not support the volatility index as a
reliable predictor of future stock returns.

(continued on next page)

31
 32

Table 1 (continued)
Market indicators Sample period Notes and formulas More references

9. Margin Account Balances:

NYSE Margin Debt 1918–2010 To put it simply, trading on margin means borrowing money from brokers to invest. We use market Fortune (2000); Glushkov
NYSE Free Credit Balances 1931–2010 aggregate margin trading statistics to gauge investors’ sentiment. Margin debt measures the total dollar (2006); Kim and Ha (2010).
NYSE Free Credit Balances on 1971–2010 amount of borrowing from brokers. Free credit balances represent the amount of money in investors’
Cash Accounts margin accounts that are free to withdraw. Furthermore, balance in a cash account reflects the money left
NYSE Free Cash Balances in after all purchases, while balance in a margin account includes cash, as well as proceeds from short sales,
Margin Accounts along with money used to meet margin requirements, and excess margin and buying power (Investopedia).
Investors trade on margin when they forecast a bullish market, therefore, the margin debt rises and the free
credit balance decreases as investors fully invest all their available funds. In contrast, during bearish
sentiment periods, the margin debt declines and the free credit balance increases, reflecting less active
trading. We use monthly released margin account statistics that monitor the overall margin trading within
the NYSE. There is controversy on if margin account statistics are contrarian indicators or not. Brown and
Cliff (2004) view margin borrowing as a bullish indicator that move in the same direction as the market, and
they find that the percentage changes of margin borrowing does not predict the market. Chen and Gu (2009)
also document that margin debt borrowers’ trading activities follow stock market trends rather than lead
the market trends. In contrast, Hirose et al. (2009) report that information about margin buying helps
1971–2010
predict future stock returns, especially for small-firm stocks at short horizons in Japanese stock market
where margin trading is dominated by individual investors.
10. Mutual Fund Balances:
USA Mutual Fund Equity 1984–2010 Investment fund statistics can reflect both informed and uninformed investors’ sentiment, as it connects Branch (1976); Edwards and
Funds Total Net Assets these two groups of market players (the uninformed fund participants and the sophisticated fund Zhang (1998); Franklin Fant
USA Mutual Fund Equity 1968–2010 managers). Theoretically, individual investors invest when they are confident with the market growth. (1999); Friesen and Sapp
Funds Cash Percentage Hence, the money inflows (outflows) measured by new sales (redemptions) of investment funds represent (2007); Randall et al. (2003);
USA Mutual Fund Equity 1984–2010 individual investors’ bullish (bearish) sentiment. They contrarily predict the market. However, the Warther (1995).
Funds Cash Percentage investment fund cash percent and liquid assets represent the level of cash, or cash equivalents, that fund
USA Mutual Fund Equity 1984–2010 managers keep for redemptions after investments. Thus, a higher cash percent represents savvy investors’
Funds New Sales bearish sentiment, indicating a market drop. We study the monthly aggregate statistics of mutual funds that
USA Mutual Fund Equity and 1954–2010 invest in the equity market only, and then examine those for mutual funds that invest in both equity and
Bond Fund Net Assets bond markets. Using data from 1933 to 1993, Neal and Wheatley (1998) find evidence that fund net
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

USA Mutual Fund Equity and 1954–2010 redemptions predict the size premiums. However Brown and Cliff (2004) find that the net redemptions do
Bond Fund Net Assets not predict market returns, neither the cash percentage by using monthly data from 1965 to 1998. Branch
USA Mutual Fund Equity and 1954–2010 (1976) also examines the cash percentages, and the author documents that some mutual funds reduce cash
Bond Fund Net Assets position before price increases while other do such during and after price increase. So that mutual funds
USA Mutual Fund Equity and 1954–2010 influence stock prices but they in turn are also influenced by stock price changes. Some studies examine the
Bond Fund Redemptions predictive ability of fund flows between different funds, like Frazzini and Lamont (2008), Ben-Rephael et al.
USA Mutual Fund Equity and 1954–2010 (2012) and Hendricks et al. (1993). We, instead, focus on the aggregate mutual fund trading statistics.
Bond Fund Redemptions
Table 1 (continued)
Market indicators Sample period Notes and formulas More references

11. Market Aggregate Number of Dividend News:

Moody’s Monthly Decreased 1956–2011 Researchers have extensively tested the fundamental information (for example, earnings information) Deshmukh et al. (2013);
Dividends contained in dividend announcements at individual stock level (for example, Nissim and Ziv, 2001; Benartzi Michaely et al. (1995).
Moody’s Monthly Extra 1956–2011 et al., 1997), as well as the sentiment information contained at this level. Baker and Wurgler (2006), for
Dividends Declared example, argue that non-dividend paying stocks are likely to be disproportionately sensitive to broad waves
Moody’s Monthly Increased 1956–2011 of investor sentiment. At the aggregate market level, previous studies like Denis and Osobov (2008) use the
Dividends Declared aggregate dollar amount of dividend to predict future returns. We, instead, look at if the aggregate number
Moody’s Monthly Omitted 1956–2011 of dividend announcements predicts the aggregate market returns. Since dividend policies largely represent
Dividends managers’ forecast on future returns (Baker and Wurgler 2004), we view them as institutional sentiment
Moody’s Monthly Resumed 1956–2011 gauges. Positive dividend news includes increased, resumed, or extra, dividends, while negative dividend
Dividends news means decreased, or omitted, dividends. The simple rationale states that if more stocks declare
positive news, a bullish market is more likely and vice versa. These data are reported in S&P’s and Moody’s
Annual Dividend Records, and to our best knowledge, they have not been studied in previous literature.
Panel B: Market Strength Indicators
1. Volume Indicators
NYSE Total Volume 1928–2011 NYSE Volume Turnover = Total Volume/Average Number of Shares Outstanding. Baker and Wurgler (2007)
NYSE Value Turnover = Total Volume in $USD/Total Market Capitalization
Total Volume Turnovers:
NYSE Share Volume 1925–2010 Technical analysis widely uses volume information to capture market liquidity and market price
Turnovera 1934–2010 movements. Baker and Stein (2004), for example, note that irrational investors are more likely to trade,
NYSE Annual Share Value which will add liquidity to the market, when they are optimistic and betting on rising stocks rather than
Turnovera when they are pessimistic and betting on falling stocks. Also, Lo et al. (2000) document that high volume
accompanying a positive price trend suggests that there may be more information content in the trend,
e.g., broader participation among investors. Moreover, Blume et al. (1994) provide evidence that volume
may provide insights regarding the quality of trader’s information that cannot be obtained from price
statistics. Scheinkman and Xiong (2003) examine the volume turnover, and find volume reveals underlying
differences of opinion, which are in turn related to valuation levels when short selling is difficult.
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

We study the raw NYSE total volume as well as two transformed volume turnovers. The turnovers adjust the
total volume to the average number of shares outstanding in the same period (note that this number may
not remain constant on a daily basis) to reflect the change on investors’ trading activities more precisely by
eliminating the possibility that increased (decreased) volumes result from more (less) shares (in terms of
volume) becoming available for trade, but not from investors’ real expectations on the market.

(continued on next page)

33
 34

Table 1 (continued)
Market indicators Sample period Notes and formulas More references

Short-Term Trading Indices:

NYSE Short-Term Trading Indexa 1965–2011 Short Term Trading Index = (No. of Advancing Issues/No. of Declining Issues)/(Up Volume/Down Volume)
NASDAQ Short-Term Trading 1972–2011 Short-term trading index studies total volumes of uptrending and downtrending stocks in a market
Indexa separately. It is sometimes referred to as the Arms Index or TRIN as first introduced by Richard W. Arms, Jr. in
1967. The short-term trading index incorporates two sorts of information; the market breadth information
and the volume information. The formula shows that a high short-term trading index results from current
lifts in the number of advancing issues alongside drops of the up volume-the current market follows a weak
uptrend with diverged volume movements-and this signals a future downward market. Conversely, a low
short-term index occurs when the number of declining issues increases, with no associated downward
market volume, so that the energy driving the downward movement is weak and a market reversal will
follow. Brown and Cliff (2004) find no merit of the index in predicting future stock returns; Wang et al.
(2006) find similar results. Whereas Simon and Wiggins (2001) find opposite results in the futures market
where the index successfully gauges investors’ bearish sentiment and predicts future returns.
2. Daily Total Market Advances and Declines:
NYSE Advances 1928–2011 Net Advancest = Advancest − Declinest Feldman (2010)
NYSE Declines 1928–2011
NYSE Net Advancesa 1928–2011 Advance–Decline Line (AD Line)t = Net Advancest + AD Linet −1 (The first value of AD Line simply takes the
value of the same period net advances)
a
NYSE AD Line 1928–2011
a Net Advancest
NYSE Percentage Net Advances 1928–2011 Percentage Net Advancest = Total Number of Stockst
NASDAQ Advances 1972–2011
NASDAQ Declines 1972–2011
NASDAQ Net Advancesa 1972–2011 This group of indicators uses the total advances/declines information to measure the strength of the
underlying trend. They are often called market breadth indicators. Total advances/declines represent the
number of stocks that experiences stock price advances/declines during a specific period. If more stocks
experience price advances, a bullish market is more likely, and vice versa. Previous literature studies two
forms of the breadth information mainly. Zakon and Pennypacker (1968) study the advance–decline line and
they find it is worthless as a leading indicator of the stock market. Brown and Cliff (2004) study the
advance/decline ratio and find it worthless in predicting the return either.
NASDAQ AD Linea 1972–2011 We study the raw advances/declines information of three major stock exchanges, the NYSE, the NASDAQ,
NASDAQ Percentage Net 1972–2011 and the Alternext. For the NYSE we also study these indicators in both daily and weekly frequencies to
Advancesa investigate whether their forecast ability differs over different horizons. We also study three popular
Alternext Advances 1959–2011 transformations of the above raw information- the net advances, the advance–decline line, and the
Alternext Declines percentage net advances. Each of the transformed indicators has unique features. The net advances use the
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Alternext Net advancesa 1959–2011 spread between advances and declines to identify the strength of the underlying trend, so that a strong
Alternext AD linea trend not only causes an increased number of advances, but this number should also exceed the number of
Alternext Percentage Net 1959–2011 declines. The advance–decline line takes the historical advance and decline information into account. If an
Advancesa underlying stock market rise is fueled only by a few stocks, and the majority of stocks either decline, or are
Weekly Total Market 1959–2011 unchanged, the rise will not last for very long.
Advances and Declines: The percentage net advances adjust the net advances for the total number of stocks in the market. More
NYSE Weekly Advances 1959–2011 stocks being available for trade gives larger potential for the differences between advances and declines (the
NYSE Weekly Declines breadth) to expand. With the development of financial markets, not surprisingly we now have a lot more
NYSE Net Advancesa issues trading within any particular stock exchange. For instance, during the 40 years from 1960 to 2000, the
1940–2010
NYSE AD Linea NYSE Issues Traded doubled from 1528 to 3083. Thus, market breadth indicators may become more accurate
1940–2010
when adjusted for changes in total stock issues within the market, so that the breadth information does not
1940–2010
purely result from the market itself, but the investors’ real views.
1940–2010
Table 1 (continued)
Market indicators Sample period Notes and formulas More references

3. Daily Total Market New

Highs and New Lows:
NYSE New Highs 1928–2011 Net New Highst = New Highst − New Lowst
NYSE New Lows 1928–2011
NYSE Net New Highsa 1932–2011 Cumulative New Highst = Net New Highst + Cumulative New Highst −1 (the first value of cumulative new
highs simply takes the value of the same period net new highs)
NYSE Cumulative Highsa 1932–2011
Net New Highst
NYSE Percentage Net New 1932–2011 Percentage Net New Highst = Total Number of Stockst
Highsa
NASDAQ New Highs 1974–2011 This group of market strength indicators generally follows the same rationale to the advances/declines
NASDAQ New Lows 1974–2011 indicators. New highs (lows) measure the number of advancing (declining) stocks reaching their 52-week
NASDAQ Net New Highsa 1974–2011 periodic highs (lows). If there are more issues reaching their periodic highs than issues reaching their
NASDAQ Cumulative Highsa 1974–2011 periodic lows, it means that a future uptrend is more likely, and vice versa. Again, similarly to the
NASDAQ Percentage Net 1974–2011 advances/declines indicators, we also include three transformed indicators to measure the market
New Highsa strength-the underlying uptrend improves when the values of the net new highs, the cumulative highs, and
the percentage net new highs increase. Brown and Cliff (2004) find that the high/low ratio does not predict
future stock returns.
Alternext New Highs 1962–2011
Alternext New Lows 1962–2011
Alternext Net New Highsa 1962–2011
Alternext Cumulative Highsa 1962–2011
Alternext Percentage Net 1963–2011
New Highsa
Weekly Total Market New Highs and New Lows:
NYSE Weekly New Highs 1937–2010
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

NYSE Weekly New Lows 1937–2010

NYSE Net New Highsa 1937–2010
NYSE Cumulative Highsa 1937–2010
a
Transformed Technical Market Indicators.
35
36 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

studies in this field largely employ just one or a few indi- 3. The total number of stocks that reach their periodic
cators as sentiment proxies to predict the market, which highs or lows. The underlying uptrend strengthens
can also lead to the data-snooping problem, since so many when more stocks advance to their periodic highs and
indicators are proposed and some receive relatively more vice versa. We also include the raw new highs/lows for
attention than others.5 Our wide range of indicators also the NYSE, NASDAQ, and Alternext and three common
avoids such risk. transformations of the raw information (net new highs,
cumulative highs, and percentage net new highs). The
2.2. Market strength indicators data for the NYSE are again available at daily and weekly
intervals.
While market sentiment indicators anticipate how in-
vestor behavior shifts market movements, market strength 3. Data and methodology
indicators measure the internal strength of these move-
ments. The fundamental goal of technical analysis is to
3.1. Sample and data
make a profit from tracking these movements, which re-
quires accurate analysis of the timing of trends, addressing
We evaluate the technical indicators’ forecastability
such questions as when does the market reach a bottom or
on the S&P 500, which proxies for the overall US stock
a peak and can it reach lower lows or higher highs? Mar-
market. The returns are calculated as the log differences
ket strength indicators answer these questions by confirm-
of current prices and prices from one period ahead. The
ing the underlying trend when the trend is strong enough
S&P 500 contains the 500 most actively traded large-cap
so that a rising or declining market may improve, even
common stocks in the US stock market. As one of the
reaching higher highs or lower lows, or by disagreeing
most historically extensive indices, the S&P 500 became
with the trend when the trend is weak and will deteriorate
available at daily, weekly and monthly frequencies in 1938,
and, thus, becomes more likely to reverse. Market strength
1918, and 1791, respectively.
is measured based on whether the majority of individual
stocks within the stock exchange participate in the uptrend We study the S&P 500 for several reasons. First, the
or the downtrend. We have 43 strength indicators, which long data series naturally shield against the potential data-
mainly use three kinds of information. snooping problem pointed out by Lakonishok and Smidt
(1988) and Lo and MacKinlay (1990). Second, we have a
1. Volume information. One of the earliest technical the- wide range of technical indicators with sufficiently long
ories, the Dow theory, documented the importance of data series designed specifically for the sophisticated US
volume: ‘‘Bull markets terminate in a period of exces- market. The 500 stocks are listed on either the NYSE or
sive activity and begin with comparatively light trans- the NASDAQ, the two largest American stock exchanges.
actions’’ (Rhea, 1932). Indicators such as the NYSE This means that the S&P 500 index correlates highly with
total volume and total volume turnovers directly mea- the NYSE and NASDAQ indices, which enables us to study
sure overall market trading activities. Moving one step
technical indicators that contain information from both of
further, the short-term trading indices use directional
these markets. Third, public investors hold the majority of
up/down volumes that track the trading volumes of ad-
the stocks in the US market. Such heavy involvement of
vancing or declining issues separately. A strong trend is
public investors satisfies the essential theoretical condition
normally accompanied by an increasing volume in the
for many of the sentiment indicators, that uninformed
same direction (Kirkpatrick and Dahlquist, 2010).
investor sentiment becomes so influential that it can shift
2. The total number of advancing/declining stocks. A trend
market movements. Last, the S&P 500 provides us with a
fueled by only a small number of stocks usually does
sufficient number of stocks to ensure considerable market
not last long. One can use raw advance/decline informa-
tion directly to measure market strength. Therefore we breadth when examining the market strength indicators.
include such raw information for the NYSE, NASDAQ, We obtain the return and indicator data from Global
and Alternext. Alternatively one can calculate the rela- Financial Data.6 Our sample frequencies vary across the
tive strength of the up/down trend through many dif- 93 indicators, with one annual indicator, 28 monthly
ferent mathematical transformations. We include the indicators, 18 weekly indicators, and 46 daily indicators
three most common transformations: the net advances, that anticipate different terms of market trends. We use
the advance/decline line, and the percentage net ad- the longest samples available for each of the indicators;
vance for the three markets above. We formulate the the annual indicator starts in 1820 and the oldest monthly,
transformation used in Table 1. Moreover, for the NYSE weekly, and daily indicators start in 1918, 1932, and 1938,
we have raw and transformed indicators data for both respectively.7 Most of our sample periods end in 2010 or
daily and weekly intervals. 2011, subject to data availability. That is, the oldest market
indicator (NYSE seat prices) has nearly 200 years of history
and the 93 indicators have an average sample length of
5 While some of the indicators in the database are discussed heavily in
54 years.
the literature, such as sentiment poll results and volatility indices, others
receive much less attention. In particular, two groups of our indicators
– exchange seat prices and the market aggregate number of dividend
6 See www.globalfinancialdata.com.
announcements – have not been studied in the previous literature to the
best of our knowledge. While we find closely related studies that can help 7 The data on market sentiment indicators II bearish percentage
explain these indicators (as discussed in Table 1), we have not found their and II bullish percentage are available from 1963, at the II website
exact application by technical analysts on our best effort; however, we www.investorsintelligence.com. We use the longest sample period from
include these indicators in our study for completeness. Global Financial Data, which starts in 1987.
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 37

As discussed in the previous section, we have several • It −1 represents periodic percentage changes of the
different indicator types: ratios (e.g., the NYSE short inter- technical indicators from one period ahead, and
est ratio), index numbers (e.g., the AAII bullish index), dol- • εt represents the residual term.
lar units (e.g., AMEX seat prices), or simply unit numbers
The methodology simply tests whether periodic variations
(such as option volumes, or the number of dividend news
of the technical indicators anticipate the next period’s
announcements). We report the mean, standard deviation,
stock market returns. The parameter β captures the rela-
minimum, and maximum values of their periodic changes
tion between market returns and the technical indicator.
in Appendix 1.
We use a conservative 10% significance level.
In addition, to test the robustness of our results, we
We also run the following regression for state-dependent
perform sub-sample analyses for each of our indicators.
predictability, following Jacobsen et al. (2007):
Since the sample periods vary greatly across indicators,
we do not define universal sub-sample periods. Instead, Rt = αt + β1 Dt −1 It −1 + β2 (1 − Dt −1 )It −1 + εt (2)
we split each indicator’s full sample into two equal sub-
where
samples. We also study state dependent predictability. We
use two sources to define business cycles independently, • Rt represents the periodic return on the S&P 500 at time
following Jacobsen et al. (2007): NBER data8 and CFNAI9 t,
data. The NBER business cycle data start in 1854, with a • It −1 represents the percentage change of the technical
monthly frequency, and we classify a year as in expansion indicator one period ahead,
(contraction) if over seven months of the year are in • Dt −1 represents a dummy variable that equals one
expanding (contracting) periods. We define each week/day (zero) during expansions (contractions) for business
as expanding (contracting) if the month of the week/day cycle analysis and one (zero) during high-sentiment
falls within an expanding (contracting) month. The CFNAI (low-sentiment) periods for sentiment cycle analysis,
data start in 1967. We classify a period as a contraction • and εt represents the residual term.
period when the CFNAI-MA3 is less than −0.7 and an The parameters β1 and β2 from Eq. (2) measure the pre-
expansion period when the CFNAI-MA3 is greater than dictability of the market indicators in expansion and con-
−0.7. Unlike the NBER data, the CFNAI data are published traction periods for business cycle analysis, respectively, or
in real time and are thus free of hindsight bias. We use the predictability in high- and low-sentiment periods for
these data to double-check our NBER results. sentiment cycle analysis, respectively. We further perform
For the sentiment regimes, we use Baker and Wurgler’s an F -test to test the statistical differences between β1 and
(2006)10 sentiment index, following Yu and Yuan (2011), β2 . We use a conservative 10% significance level and apply
to define high-/low-sentiment regimes. We classify a year White’s standard error corrections on all t-statistics and
as a high-sentiment (low-sentiment) year if the prior year χ -statistics to counter heteroskedasticity issues.
has a positive (negative) value of the index. Baker and
Wurgler (2006) calculate the index as the first principle
4. Empirical results
component of six measures of investor sentiment, which
are the closed-end fund discount, the NYSE share turnover,
the number of IPOs, the average first-day return of IPOs, the 4.1. Main OLS results
equity share in new issues, and the dividend premium. The
first principle calculation eliminates noise and captures the Table 2 presents our main OLS results for the full sample
common component of the different sentiment measures. in the first three columns, followed by the OLS results for
Furthermore, the authors first regress the six sentiment two equal-fold sub-samples. For each sample, we report
measures on a set of macroeconomic variables to remove the sample periods, β estimates, and White standard error-
business cycle information and then use the residuals as corrected t-statistics. Panels A and B present the results
input for first principle component analysis. Therefore our for the market sentiment and market strength indicators,
sentiment time varying analysis does not overlap with the respectively.
business cycle-varying analysis. The full-sample results show that 30 out of the total
93 market indicators predict the market at the 10% sig-
nificance level. This includes 18 market sentiment indica-
3.2. Methodology
tors and 12 market strength indicators. In regard to the
different underlying information, five groups of indica-
We run standard OLS regression to test the predictabil-
tors (option volumes, Barron’s confidence index, exchange
ity of each of the 93 technical indicators:
seat prices, total volumes, and total volume turnovers)
R t = α t + β I t − 1 + εt (1) exhibit no predictability at all. On the other hand, short
sales volumes, volatility indices, and raw total market ad-
where
vance/decline indicators seem to perform better than the
• Rt represents the daily/weekly/monthly/annual log other indicators at first glance; all of them show (marginal)
returns of the S&P 500 index, significance in predicting the market, except the NASDAQ
100 volatility index.
Although 30 indicators show some preliminary
8 See http://www.nber.org/cycles.html. predictability, we also need to consider an important and
9 See http://www.chicagofed.org/webpages/publications/cfnai/. relevant question: Do they work in the way that techni-
10 See http://people.stern.nyu.edu/jwurgler/. cal theory expects? In other words, can one really make a
38 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Table 2
OLS results.
Market indicators Full sample Sub-sample 1 Sub-sample 2
Period β (∗10−3 ) t-stats Period 1 β (∗10−3 ) t-stats Period 2 β (∗10−3 ) t-stats

Panel A: Market Sentiment Indicators

Option Volumes:
CBOE Calls Volume 1989–2011 −0.41 −1.17 1989–1999 −0.85 −1.55 2000–2011 −0.23 −0.55
CBOE Puts Volume 1989–2011 −0.01 −0.12 1989–1999 −0.02 −0.52 2000–2011 0.07 0.20
OEX Calls Volume 1989–2011 0.00 −1.26 1989–1999 0.00 −1.51 2000–2011 0.00 −1.29
OEX Puts Volume 1989–2011 0.00 0.25 1989–1999 0.00 0.16 2000–2011 0.00 0.34
CBOE Ratio of Traded Value of 1986–2011 0.63 0.77 1986–1998 −0.07 −0.04 1999–2011 1.11 1.36
Puts to Calls
Odd-Lots Volumes:
NYSE Odd Lot Purchases 1970–2011 0.00 −4.90 1970–1990 1.07 1.38 1991–2011 0.00 −4.05
NYSE Odd Lot Sales 1970–2011 0.11 0.27 1970–1990 0.78 1.25 1991–2011 −0.05 −0.10
NYSE Odd Lot Shorts 1970–2011 0.00 1.14 1970–1990 −0.04 −0.48 1991–2011 0.00 1.18
Short Sales Volumes:
NYSE Short Sales-Members 1940–2008 6.68 7.15 1940–1974 6.76 4.85 1975–2008 6.64 5.26
NYSE Short Sales-General 1940–2008 2.63 2.58 1940–1974 0.93 0.97 1975–2008 7.18 3.06
Public
NYSE Short Sales-Specialists 1940–2008 5.90 5.82 1940–1974 6.99 4.93 1975–2008 5.30 3.77
NYSE Short Sales-Total 1940–2008 6.80 5.59 1940–1974 5.74 4.04 1975–2008 7.88 4.27

Short Interests:
NYSE Short Interest Ratio 1931–2010 −23.19 −2.22 1931–1970 −23.16 −2.22 1971–2010 −23.19 −2.22
NYSE Short Interest Shares 1931–2010 −2.93 −0.12 1931–1970 −2.92 −0.12 1971–2010 −2.93 −0.12
AAII/II Sentiment Indices:
AAII Bearish Index 1989–2010 0.02 0.01 1989–1999 −1.02 −0.38 2000–2010 0.77 0.31
AAII Bullish Index 1989–2010 6.39 2.26 1989–1999 5.17 1.55 2000–2010 7.50 1.66
AAII Neutral Index 1989–2010 −8.70 −2.83 1989–1999 −3.05 −0.92 2000–2010 −10.75 −2.65
Investors Intelligence Bearish 1987–2010 −1.04 −0.11 1987–1998 −5.17 −0.44 1999–2010 1.19 0.09
Percentage
Investors Intelligence Bullish 1987–2010 −0.36 −0.03 1987–1998 8.19 0.74 1999–2010 −9.74 −0.50
Percentage
Confidence Index:
Barron’s Confidence Index 1932–2010 36.44 0.78 1932–1970 43.62 0.73 1971–2010 25.05 0.33
Exchange Seat Prices:
AMEX Seat Prices 1921–1993 3.38 0.48 1921–1958 12.74 0.82 1959–1993 −3.32 −0.76
Volatility Indices:
CBOE S&P 500 Volatility Index 1986–2011 7.01 1.74 1986–1998 3.54 0.50 1999–2011 13.10 3.04
CBOE NASDAQ Volatility Index 2001–2011 13.28 2.10 2001–2005 5.21 0.74 2006–2011 17.22 1.99
CBOE S&P 100 Volatility Index 1986–2011 7.33 1.96 1986–1998 4.09 0.60 1999–2011 12.16 3.14
AMEX NYSE Arca NASDAQ 100 2001–2011 4.00 0.61 2001–2005 0.11 0.02 2006–2011 5.61 0.62
Volatility Index
CBOE DJIA Volatility Index 2005–2011 13.39 1.90 2005–2007 11.24 2.52 2008–2011 15.75 1.13
Margin Account Balances:
NYSE Margin Debt 1918–2010 −0.72 −0.02 1918–1963 2.05 0.05 1964–2010 −12.29 −0.24
NYSE Free Credit Balances 1931–2010 80.49 2.11 1931–1970 120.88 2.07 1971–2010 48.63 0.95
NYSE Free Credit Balances on 1971–2010 22.34 0.63 1971–1990 −5.56 −0.11 1991–2010 57.21 1.06
Cash Accounts
NYSE Free Cash Balances in 1971–2010 1.66 0.04 1971–1990 −30.90 −1.00 1991–2010 88.08 1.12
Margin Accounts
Mutual Fund Balances:
USA Mutual Fund Equity Funds 1984–2010 92.74 1.44 1984–1996 18.63 0.19 1997–2010 129.87 1.41
Total Net Assets
USA Mutual Fund Equity Funds 1968–2010 −20.76 −0.69 1968–1988 −1.97 −0.06 1989–2010 −59.02 −0.99
Cash Percentage
USA Mutual Fund Equity Funds 1984–2010 −4.74 −2.89 1984–1996 −5.73 −5.22 1997–2010 6.10 0.25
Redemptions
USA Mutual Fund Equity Funds 1984–2010 6.59 0.54 1984–1996 5.99 0.44 1997–2010 4.21 0.17
New Sales
USA Mutual Fund Equity and 1954–2010 10.50 6.14 1954–1981 91.76 1.32 1982–2010 9.76 8.05
Bond Fund Net Assets
USA Mutual Fund Equity and 1954–2010 −17.87 −0.78 1954–1981 −2.94 −0.10 1982–2010 −34.84 −0.91
Bond Fund Cash Percent
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 39

Table 2 (continued)

Market indicators Full sample Sub-sample 1 Sub-sample 2

Period β (∗10−3 ) t-stats Period 1 β (∗10−3 ) t-stats Period 2 β (∗10−3 ) t-stats

USA Mutual Fund Equity and 1954–2010 13.26 0.51 1954–1981 26.58 0.86 1982–2010 −5.70 −0.13
Bond Fund Liquid Assets
USA Mutual Fund Equity and 1954–2010 −10.50 −0.91 1954–1981 −13.13 −0.99 1982–2010 −8.53 −0.46
Bond Fund Redemptions
USA Mutual Fund Equity and 1954–2010 7.89 0.85 1954–1981 7.28 0.68 1982–2010 8.73 0.53
Bond Fund New Sales
Number of Dividend News:
Moody’s Monthly Decreased 1956–2011 40.61 1.57 1956–1984 90.97 3.22 1985–2011 −35.03 −0.73
Dividends
Moody’s Monthly Extra 1956–2011 −63.32 −1.38 1956–1984 −161.23 −1.91 1985–2011 −41.05 −0.80
Dividends Declared
Moody’s Monthly Increased 1956–2011 −97.86 −1.97 1956–1984 −125.17 −2.17 1985–2011 −57.17 −0.66
Dividends Declared
Moody’s Monthly Omitted 1956–2011 7.60 0.24 1956–1984 35.53 0.99 1985–2011 −40.49 −0.72
Dividends
Moody’s Monthly Resumed 1956–2011 15.28 0.81 1956–1984 73.86 2.09 1985–2011 −1.88 −0.07
Dividends
S&P Monthly Dividend 1955–2010 0.43 0.45 1955–1982 1.37 1.70 1983–2010 −1.09 −0.53
Decreases Declared
S&P Monthly Extra Dividends 1955–2010 4.48 2.17 1955–1982 5.27 2.23 1983–2010 3.60 0.99
Declared
S&P Monthly Increased 1955–2010 2.11 0.57 1955–1982 8.39 1.39 1983–2010 −0.79 −0.20
Dividends Declared
S&P Monthly Omitted 1955–2010 0.88 0.68 1955–1982 0.22 0.10 1983–2010 1.25 0.91
Dividends Declared
S&P Monthly Resumed 1955–2010 2.85 1.89 1955–1982 3.29 2.18 1983–2010 2.39 1.02
Dividends Declared
Panel B: Market Strength Indicators
Total Volume:
NYSE Total Volume 1928–2011 0.09 0.83 1928–1969 0.07 0.71 1970–2011 0.65 3.48

Total Volume Turnovers:

NYSE Share Volume Turnover 1925–2010 5.39 0.13 1925–1967 84.20 1.72 1968–2010 −70.03 −1.27
NYSE Annual Share Value 1934–2010 28.23 0.64 1934–1971 40.38 0.85 1972–2010 −43.56 −0.37
Turnover
Short-Term Trading Indices:
NYSE Short-Term Trading 1965–2011 −0.49 −2.15 1965–1987 −1.12 −2.94 1988–2011 0.07 0.24
Index
NASDAQ Short-Term Trading 1972–2011 −0.01 −1.16 1972–1991 −0.18 −1.18 1992–2011 −0.01 −1.15
Index
Daily Total Market Advances
and Declines:
NYSE Advances 1928–2011 0.51 2.98 1928–1969 0.43 2.12 1970–2011 0.77 2.44
NYSE Declines 1928–2011 −0.72 −3.65 1928–1969 −0.53 −2.36 1970–2011 −1.18 −3.19
NYSE Net Advances 1928–2011 0.00 0.49 1928–1969 0.00 −0.33 1970–2011 0.00 0.99
NYSE AD Line 1928–2011 0.00 −0.35 1928–1969 0.00 −0.41 1970–2011 0.00 −0.12
NYSE Percentage Net Advances 1940–2011 0.00 0.36 1928–1969 0.00 −0.51 1970–2011 0.00 0.99

NASDAQ Advances 1972–2011 0.23 1.48 1972–1991 0.35 2.34 1992–2011 0.00 −0.01
NASDAQ Declines 1972–2011 −0.10 −3.41 1972–1991 −0.09 −4.65 1992–2011 −0.76 −1.09
NASDAQ Net Advances 1972–2011 0.00 −0.50 1972–1991 0.01 0.55 1992–2011 0.00 −0.97
NASDAQ AD Line 1972–2011 0.00 −0.22 1972–1991 0.00 −0.53 1992–2011 0.00 0.16
NASDAQ Percentage Net 1972–2011 0.00 −0.51 1972–1991 0.01 0.55 1992–2011 0.00 −0.97
Advances
Alternext Advances 1959–2011 1.18 4.02 1959–1984 1.53 5.38 1985–2011 0.47 0.65
Alternext Declines 1959–2011 −1.04 −2.46 1959–1984 −1.06 −1.85 1985–2011 −0.98 −2.03
Alternext Net Advances 1959–2011 0.01 0.80 1959–1984 0.00 0.74 1985–2011 0.01 0.50
Alternext AD Line 1959–2011 0.00 −0.03 1959–1984 0.00 −0.14 1985–2011 0.00 0.07
Alternext Percentage Net 1959–2011 0.01 0.60 1963–1986 0.00 −0.31 1987–2011 0.01 0.87
Advances
Weekly Total Market Advances
and Declines:
NYSE Weekly Advances 1940–2010 −1.49 −3.33 1940–1974 −1.14 −1.50 1975–2010 −1.69 −3.38
NYSE Weekly Declines 1940–2010 0.65 1.21 1940–1974 −0.28 −0.39 1975–2010 1.75 2.05
(continued on next page)
40 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Table 2 (continued)

Market indicators Full sample Sub-sample 1 Sub-sample 2

Period β (∗10−3 ) t-stats Period 1 β (∗10−3 ) t-stats Period 2 β (∗10−3 ) t-stats

NYSE Net Advances 1940–2010 0.00 0.22 1940–1974 0.00 −0.17 1975–2010 0.00 0.35
NYSE AD Line 1940–2010 −1.20 −0.52 1940–1974 −1.16 −0.49 1975–2010 −47.20 −0.85
Daily Total Market New Highs
and New Lows:
NYSE New Highs 1928–2011 0.14 3.61 1932–1971 0.71 4.86 1972–2011 0.10 9.80
NYSE New Lows 1932–2011 −0.13 −1.50 1932–1971 −0.12 −1.23 1972–2011 −0.15 −0.93
NYSE Net New Highs 1932–2011 0.04 1.77 1932–1971 0.05 1.44 1972–2011 0.03 1.37

NYSE Cumulative Highs 1932–2011 −0.01 −0.34 1932–1971 0.00 −0.09 1972–2011 −0.01 −0.30
NYSE Percentage Net New 1932–2011 0.04 1.60 1932–1971 0.05 1.04 1972–2011 0.03 1.37
Highs
NASDAQ New Highs 1974–2011 −0.16 −0.43 1974–1992 0.63 1.67 1993–2011 −0.71 −1.20
NASDAQ New Lows 1974–2011 0.25 1.26 1974–1992 −0.24 −1.01 1993–2011 0.67 2.23
NASDAQ Net New Highs 1974–2011 −0.01 −0.21 1974–1992 −0.03 −0.54 1993–2011 0.01 0.21
NASDAQ Cumulative Highs 1974–2011 0.03 0.98 1974–1992 0.01 0.21 1993–2011 0.04 1.20
NASDAQ Percentage Net New 1974–2011 −0.01 −0.22 1974–1992 −0.03 −0.55 1993–2011 0.01 0.19
Highs

Alternext New Highs 1962–2011 0.20 2.23 1962–1986 0.14 2.40 1987–2011 0.40 1.69
Alternext New Lows 1962–2011 −0.06 −0.89 1962–1986 −0.27 −2.31 1987–2011 0.01 0.09
Alternext Net New Highs 1962–2011 0.00 0.11 1962–1986 −0.05 −0.98 1987–2011 0.04 0.79
Alternext Cumulative Highs 1962–2011 −0.03 −0.88 1962–1986 −0.03 −0.67 1987–2011 −0.04 −0.70
Alternext Percentage Net New 1962–2011 0.01 0.30 1963–1986 −0.03 −0.54 1987–2011 0.04 0.71
Highs
Weekly Total Market New Highs
and New Lows:
NYSE Weekly New Highs 1937–2010 0.11 0.26 1937–1973 0.17 0.32 1974–2010 0.05 0.07
NYSE Weekly New Lows 1937–2010 −0.30 −0.74 1937–1973 −0.33 −0.76 1974–2010 −0.09 −0.08
NYSE Net New Highs 1937–2010 0.11 1.88 1937–1973 0.12 1.78 1974–2010 0.10 1.32
NYSE Cumulative Highs 1937–2010 −0.01 −3.62 1937–1973 0.54 2.19 1974–2010 −0.01 −3.55
This table reports the OLS results of the regression model Rt = αt + β It −1 + εt for full samples and two equal length sub-samples. Rt represents S&P 500
periodic returns calculated as log differences of the S&P 500 Index values, It −1 represents periodic percentage changes of market indicators. We obtain all
data from the Global Financial Data. The t-statistics reported are White standard errors corrected and marked in bold if significant at 10% significance level.
Panel A and Panel B report results for market sentiment and market strength indicators respectively.

profit from following the technical textbook? Our results the predictability depends on whether raw or trans-
provide a mixed answer to this question, with 10 of the 30 formed information is used. Interestingly, at both daily and
market indicators showing significant predictability, but weekly frequencies, all eight raw advance/decline indica-
with signs opposite from the expected. Eight sentiment tors (marginally) predict the market, in contrast with none
indicators predict the market differently from what the- of the transformed indicators. Hence, transformation of
ory implies. Typical contrarian indicators such as the NYSE advance/decline information does not provide any further
short interest ratio, the AAII bullish index, and US mutual insight into market trends, it even appears to cause infor-
fund equity fund redemptions do not actually exhibit a mation loss. Furthermore, different underlying predictive
contrarian nature. Instead, they capture the correct mar- horizons can incur variations in a particular indicator’s pre-
ket direction. Hence, the traditional market wisdom that dictability. At daily frequencies, raw advances predict the
trading against uninformed investors no longer seems to market positively and raw declines predict the market neg-
hold here. In contrast, indicators on savvy investors, such atively, which is in line with theory. However, such rela-
as NYSE members/specialists, who are supposed to be cor- tions reverse at weekly frequencies. In addition, the new
rect, are also unreliable. For example, the increase in savvy highs/new lows indicators work better in their raw forms
short sales should predict a downward market. However, at daily frequencies, whereas they only work in their trans-
our results show that it is actually associated with a fu- formed forms at weekly frequencies.
ture market rise. Similarly, two market strength indicators, The mixed full-sample results make it difficult to con-
weekly NYSE advances and weekly NYSE cumulative highs, clude yet whether the market indicators are useful or not,
which should predict the market positively, actually have with 30 out of the 93 indicators showing some prelimi-
negative signs. Such results imply that, even though these nary predictability, especially considering that they pro-
market indicators show a significant relation with future vide different indications than expected by the underlying
returns, trading on them in the way indicated by theory theory and the predictability can differ with the method
will incur losses. of using the information. We then further test the gen-
In addition, the predictability of the same market eral stability of the indication by splitting the full samples
strength information varies with the way it is used. First, into two sub-samples of equal length. Note that since the
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 41

original full-sample lengths vary across indicators, the two 4.2. Rolling window regressions
sub-samples also have different lengths for different indi-
cators. We run a 10-year11 rolling window regression on the
Our sub-sample results cast further doubt on the pre- 10 indicators that survive the sub-sample test. For each
dictive power of the market indicators. Only 10 market indicator, we first run the original OLS regression on the
indicators remain predictive in both of the sub-samples: first 10 years of the sample and then move the sample one
month forward by replacing the observations in the first
three different types of investors’ short-sales volumes, the
month of the previous 10 years with those of the latest
NYSE short interest ratio, daily NYSE advances and de-
one month and repeat the OLS regression. Thus, the new
clines, daily Alternext declines, daily new highs of the NYSE
regression window remains 10 years but rolls one month
and Alternext, and weekly NYSE cumulative highs. We
forward. We repeat this process until the last observation
highlight the 10 predictive market indicators in boxes. If in the full sample is included in the last regression. The
we further group these indicators by their underlying in- observed β should maintain at a relatively stable level if
formation, only some short sales statistics and market ad- the indicator predicts the market consistently over time.
vances/declines or new highs/lows information may still Fig. 1 plots the rolling OLS β values in solid lines
contain some predictive value. and their 90% confidence bounds in dotted lines over
The sub-sample results have several additional implica- time. On average, all three of the NYSE short sales-
tions. First, Branch (1976) suggests that the predictability members/specialists/total maintain reasonable consis-
of technical indicators may disappear over time, since they tency in predicting the market, except for the short period
will attract more investors to exploit their predictability in 2001, when the market was closed because of the 911
after they are found to work. In this case, we should find attack. Surprisingly, members’ and specialists’ short sales
more efficient technical indicators in the first sub-sample. are persistently positively related to the market over time.
However, we discover similar numbers of efficient tech- That is, when the informed NYSE members, or the special-
ists, increase their short positions to hedge against a mar-
nical indicators in the first and second sub-samples, with
ket fall, the market actually rises. Besides, we discover an
only 10 indicators actually showing statistical significance
intriguing pattern for the short interest ratio: The sign of β
in the latter half of the sample period. Nevertheless, most
keeps switching between positive and negative over time.
market indicators (52 out of 93) show no predictability in
This casts strong doubts on its predictability, since it seems
either sub-sample; it appears that these 52 indicators have difficult to follow the varying indication it provides from
never worked across their full history, which can be as long time to time. In addition, the wide confidence bounds make
as 193 years (the sample of NYSE seat prices starts in 1820). it even harder to rely on such predictability.
It seems that the argument that predictability is gradually For the market strength indicators, the five raw indi-
exploited over time does not hold. cators (NYSE advances/declines/new highs and Alternext
Second, one may argue that the development of finan- declines/new highs) generally predict the market consis-
cial markets enabled some trading methods that masked tently over time, although at the same time, except for
the true informational content of some historically useful NYSE declines, the market indicators often experience pe-
technical indicators and led to the loss of their predictabil- riods with relatively wide confidence bounds for the β es-
ity. For example, Kirkpatrick and Dahlquist (2010) argue timates. For example, NYSE advances have a wider confi-
that margin debt, which was previously a very reliable in- dence bound from late 1947 to late 1958 and from early
dicator, is no longer an accurate gauge of investor sen- 1987 to late 1997. The same is also the case for NYSE new
timent because investors can, through using derivatives, highs after the 2008 financial crisis period. In contrast, the
large fluctuation of β largely eliminates the NYSE weekly
hold positions outside the Federal Reserve requirements
cumulative highs as reliable market predictors. The rolling
for margins. Our results, however, do not appear to support
window regression shows that for NYSE weekly cumula-
such an argument, since margin debt does not work in the
tive highs, β is positive before 1974 (positive and close to
first sub-sample before 1963, when stock index derivatives
zero from 1947 to 1953), when it switches sign and re-
were not as widely used by the public as they are now. We mains negative and close to zero for the following 10 years
actually also find no predictability for the rest of the margin to 1984. Then its sign switches again to be positive un-
account statistics indicators. This further supports the view til 1997, whereafter it becomes almost always negative.
that predictability does not seem to decrease over time but, This probably explains why, in the sub-sample test, NYSE
rather, probably to a large extent never existed. weekly cumulative highs predict the market differently in
Of our 93 market indicators, only 10 survive the sub- the two sub-samples.
sample analysis. It should be noted that the weekly NYSE
cumulative highs predict the market differently in the two
11 One could argue that a regression window of 10 years is not adequate.
sub-sample periods, positively in the first sub-sample and
For example, Jacobsen and Dannenburg (2003) suggest that, for monthly
negatively in the second. This raises an intriguing question:
observations, 50 years of data are required to produce reliable GARCH
Even though the 10 market indicators are overall predic- estimates. However, in our case we use the longest sample available for
tors of the market in the long run, do the indications they each indicator and some of these indicators have a full history of only
supply remain the same over time? How stable is the pa- around 50 years (e.g., Alternext new highs). Our primary focus also lies in
the stability of predictability across time and not the exact magnitude of
rameter β ? We perform rolling window regressions to an- the β value; a 10-year window for our rolling window regressions should
swer these questions. serve such a goal.
42 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Fig. 1. 10-year rolling window OLS regressions.

 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 43

The rolling window regression results warn us about are independent and identically distributed normal (see
the danger of using short interest ratios and NYSE cumu- Table 3).
lative highs as market predictors, even though they all The last four columns report the α and β estimates
exhibit statistical significance in the full sample and sub- and their associated t-values for Jensen’s α estimation. We
samples on first examination. This again emphasizes the estimate Jensen’s α using the regression
importance of using a long sample period. The other eight p f f
indicators generally present relatively stable predictabil- rt − rt = α + β(rtm − rt ) + εt
ity, although to some degree they are exposed to the p f
problem of wide confidence bounds. We perform several where rt , rt , and rtm represent the returns of the technical
robustness checks to address this problem in Section 6 and trading strategies, risk-free rate, and market returns,
the results remain largely the same. respectively. The term α then captures the excess return
on a given systematic risk level β of the technical trading
strategy by using the buy and hold strategy as the
4.3. Economic significance
benchmark.
Compared with the buy and hold strategy, the techni-
Our last step takes into account transaction costs and
cal strategies generally have both lower returns and lower
examines the risk-adjusted returns of investing on the
eight indicators that provide relatively reliable indication risks. Three technical strategies (NYSE advances, declines,
over time. We use the methodology of Driesprong et al. and new highs) even have negative returns, on average,
(2008) to test the economic significance of the market indi- before considering risks, which suggests investing on risk-
cators. For each of the eight market indicators, we calculate free assets will be more mean–variance efficient. Further-
our portfolio return by using OLS estimates, as follows: more, in terms of the Sharpe ratio that measures the price
for each unit of risk, none of the technical strategies sig-
• We first split the sample into two equal lengths and we nificantly outperforms the buy and hold strategy. In fact,
estimate the OLS model parameters αt and βt using the most technical strategies have negative Sharpe ratios that
first half of our sample. underperform investing on risk-free assets. The results of
• At time t + 1, we use αt , βt , and the last market Jensen’s α estimation provide more or less similar implica-
indicator change It to calculate the expected return tions. Although all the β estimates are significantly below
E (Rt +1 ). Then we compare E (Rt +1 ) with the same one, indicating lower risk levels, none of the technical trad-
f
period’s risk-free rate rt +1 .12 We fully invest in the ing strategies produce a more positive excess return, cap-
f
market if E (Rt +1 ) is higher than rt +1 , so that the tured by α , than the buy and hold strategy at this level of
portfolio return rtP+1 = rtm+1 , and we fully invest in risk- risk. Overall, our OLS results indicate none of the 93 market
f f indicators generate returns outperforming the market.
free assets if E (Rt +1 ) is lower than rt +1 ; thus rtP+1 = rt +1 .
• We re-estimate our model every period to update the
model whenever a new observation becomes available 5. Time-varying predictability
and then calculate our portfolio returns.
• Similarly to Driesprong et al. (2008), we assume a Our conclusion may be too restrictive yet if return
switching cost of 0.10% between the market and risk- predictability is state dependent. Prior literature has
free assets, in accordance with Solnik (1993). shown that some return predictability models’ effective-
We then compare the risk and return pattern of our ness varies across business cycles (Dangl and Halling, 2012;
portfolio with that of a naive buy and hold portfolio; we Henkel et al., 2011; Jacobsen et al., 2007) or across senti-
document the results in Table 2. We first report the mean, ment regimes (Stambaugh et al., 2012). Several of our mar-
standard deviation, and Sharpe ratio for the buy and hold ket indicators exhibit sign-switching predictability across
strategy and columns 7–9 report those of the technical time, for example, the short interest ratio and weekly NYSE
strategy. We calculate the Sharpe ratio as cumulative highs. If some of the technical market indica-
tors have time-varying or state-dependent predictability,
p f p
Sharpe ratio = (rt − rt )/σt it remains possible that they have not been picked out by
p
where rt represents the returns of the technical trading our full-sample and sub-sample tests. Hence, in this sec-
f tion, we implicitly investigate the time variation and state
strategies; rt represents the risk-free rate, which equals
dependency of the 93 indicators.
the US three-month Treasury bill rates; and σ p represents
p
the standard deviation of rt . The next column reports the
t-values, testing the null hypothesis that the Sharpe ratio 5.1. Business cycle-varying predictability
of the buy and hold strategy equals that of the technical
strategy. The significance test is performed according to Han et al. (2013) find that the moving average trading
the methodology proposed by Lo (2002) and De Roon et al. strategies generate much higher abnormal returns in re-
p f
(2012), which assumes that the excess returns rt − rt cessions. Chordia and Shivakumar (2002) discover similar
evidence for momentum strategies, which generate pos-
itive returns only during expansions. On the other hand,
12 We source our risk-free rate data from Global Financial Data using Griffin et al. (2003) claim profitable momentum strategies
three-month US Treasury bill rates. in both good and bad economic states. All these studies
 44

Table 3
OLS economic significance test.
Market indicators Frequency N Buy and hold strategy Technical strategy
Mean Std. Dev. Sharpe ratio Mean Std. Dev. Sharpe ratio t-stats α (∗10−3 ) t-stats β t-stats
(∗10−3 ) (∗10−2 ) (∗10−2 ) (∗10−3 ) (∗10−2 ) (∗10−2 )
Panel A: Market Sentiment Indicators
NYSE Short Weekly 1 743 1.28 2.27 0.84 0.93 1.70 −0.97 1.06 −0.27 −0.97 0.57 16.34
Sales-Members
NYSE Short Weekly 1 742 1.28 2.27 0.84 1.03 1.62 −0.42 0.69 −0.16 −0.57 0.51 13.69
Sales-Specialists
NYSE Short Sales-Total Weekly 1 742 1.28 2.27 0.84 1.13 1.69 0.18 0.38 −0.06 −0.23 0.56 15.76
Panel B: Market Strength Indicators
NYSE Advances Daily 11 024 0.24 1.06 0.85 −0.08 0.69 −3.35 5.15 −0.28 −5.51 0.40 17.23
NYSE Declines Daily 11 024 0.24 1.06 0.85 −0.05 0.84 −2.36 5.13 −0.26 −5.25 0.62 20.14
Alternext Declines Daily 6 608 0.30 1.18 1.57 −0.07 0.99 −1.87 4.79 −0.32 −4.45 0.69 15.48
NYSE New Highs Daily 10 307 0.26 1.07 1.07 0.15 0.98 −0.02 2.52 −0.10 −2.38 0.82 62.32
Alternext New Highs Daily 6 109 0.26 1.20 1.24 0.14 1.15 0.26 2.78 −0.11 −2.67 0.93 62.96
This table compares risk and return characteristics of buy and hold strategies with strategies based on market indicators that show significant predictability in both sub-samples under the OLS regressions. We
report means, standard deviations and Sharpe ratios for buy and hold strategies and technical strategies consequently, then we report t-statistics testing the null hypothesis that the Sharpe ratios of the two
strategies are equal. We then report Jensen’s α estimation results for technical strategies in the last four columns. α values indicate excess returns generated by the technical strategies at given risk level β
over market returns. We also present associated t-statistics testing their differences from zero for α and β values. All t-statistics are White standard errors corrected and marked in bold if significant at 10%
significance level.
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 45

provide evidence for business cycle-related predictability 5.2. Sentiment regime-varying predictability
but do not pay attention to the market indicators. We seek
to fill the gap here for the effect of market indicators on We also test a second set of possible predictability
business cycle-varying predictability, if any. regimes: the sentiment regimes introduced by Yu and
We use the monthly NBER business cycle data13 that Yuan (2011). These authors find a significantly positive
start in 1854 to define expansion and contraction peri- mean–variance relationship during low-sentiment peri-
ods.14 We report the regression results in Table 4. The first ods but no relation during high-sentiment periods in
two columns repeat our full-sample results again for com- which sentiment shifts price away from its fundamen-
parison. Columns 3–6 report β1 and β2 , which measure tal values. Stambaugh et al. (2012) also document that
the predictability of the market indicators in expansions a set of asset pricing anomalies becomes stronger dur-
ing high-sentiment periods. Their finding could have an
and contractions, with White standard error-corrected t-
impact on the predictability of market indicators. Many
statistics. The last column reports the F -test results, testing
market indicators work on the basis of measuring in-
the statistical differences between β1 and β2 .
vestor sentiment, which technical analysis believes is the
Generally, market indicators’ predictability does not
force that drives prices from their fundamental values
seem to strengthen under different business states. We (Kirkpatrick and Dahlquist, 2010). We therefore won-
have 26 predictive indicators in expansions and 21 in con- der whether these market indicators show stronger pre-
tractions, suggesting overall market indicators do not seem dictability during high-sentiment periods in which such
to work better in one business state. In addition, com- forces become stronger. If this is the case, the full-sample
pared with the 30 significant results discovered under the analysis can miss such predictability.
full sample, our results seem to suggest that predictabil- Following Yu and Yuan (2011), we use the annual Baker
ity is not strengthened even if we allow it to be time and Wurgler (2006) sentiment index to define sentiment
varying across business cycles. The F -test results give a periods. Baker and Wurgler calculate a composite senti-
similar message, that the predictability of most market ment index as the first principle component of six mea-
indicators (74 out of the total 93) in contractions is not sures of investor sentiment, namely, the closed-end fund
statistically different from that in expansions. This largely discount, the NYSE share turnover, the number of IPOs,
eliminates the possibility that these indicators work better the average first-day return of IPOs, the equity share in
in one business state and not the other, offsetting overall new issues, and the dividend premium. The first princi-
predictability. ple calculation eliminates noise and captures the common
To allow the maximum benefit of the doubt, we component of the different sentiment measures. Further-
perform economic significance tests for indicators that more, the authors first regress the six sentiment measures
exhibit significant predictability in any of the expansion on a set of macroeconomic variables to remove business
or contraction periods and we tabulate the results in cycle information and they then use the residuals as in-
Tables 5 and 6, respectively. Generally, the technical put for first principle component analysis. Therefore, our
trading strategies have lower risk levels than the buy sentiment time-varying analysis does not overlap with the
and hold strategies in both business states, although business cycle-varying analysis above. We then classify a
year as a high-sentiment year if the prior year has a pos-
they largely do not beat the buy and hold strategies in
itive Baker and Wurgler (2006) index value. We use the
returns for each unit of risk as measured by the Sharpe
same regime-switching methodology as above and define
ratio. Jensen’s α results largely tell the same story: The
the dummy variable as equal to one (zero) during high-
technical trading strategies usually have low β levels,
sentiment (low-sentiment) periods.
which means low systematic risk, but they do not generate
We present the sentiment regime-varying results in
excess returns to the market at the risk level β either.
Table 7.15 Again, we first recall the full-sample results in
We have two exceptions in the expansion periods. Equity
the first two columns and then subsequently present the
fund redemptions have a Sharpe ratio significantly higher results during high- and low-sentiment periods; lastly, we
than the market’s and a marginally significant Jensen’s present the F -test results, testing the differences between
α and NYSE net new highs have a significant positive high- and low-sentiment periods.
Jensen’s α and a marginally significantly higher Sharpe We find a total of 21 and 25 market indicators predict-
ratio. These two indicators may show some predictability ing the market during high- and low-sentiment periods,
during expansion periods only; however, this finding does respectively. Contrary to what was expected, we do not
not alter our main conclusions. discover more predictive indicators during high-sentiment

13 See http://www.nber.org/cycles.html. 15 Notice that the full-sample results are for the longest sample available
14 Our indicators have different frequencies and we define expansions for each of the indicators, while the sentiment regime varying results
and contractions as follows: The only annual indicator, NYSE seat prices, are for the period from 1967 to 2011 where the sentiment index is
starts in 1820, whereas, due to data availability on business cycles, available. This may cause some unusual effect during comparison. For
our time-varying evaluation on the annual indicator starts in 1854. We example, the NYSE Total Volume predicts returns in both regimes but
classify a year as in expansion (contraction) if over seven months of the not the full-sample. However this would not affect our main conclusion
year are in expanding (contracting) periods. We define each week/day since we perform further analysis for the best benefit of doubt, also those
as in expansion (contraction) if the month it falls in is expanding indicators whose sample starts after 1967 would not suffer from this
(contracting). problem.
46 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Table 4
NBER business cycle time-varying results.
Market indicators Full sample Expansions Contractions Chi-statistic
β(∗10−3 ) t-stats β1 (∗10−3 ) t-stats β2 (∗10−3 ) t-stats

Panel A: Market Sentiment Indicators

Option Volumes:
CBOE Calls Volume 0.00 1.15 −0.31 −0.88 −3.22 −0.90 0.65
CBOE Puts Volume −0.01 −0.12 0.01 0.18 −0.77 −0.25 0.06
OEX Calls Volume 0.00 −1.26 0.00 −1.17 −1.72 −1.31 1.73
OEX Puts Volume 0.00 0.25 0.00 0.27 1.55 1.18 1.40
CBOE Ratio of Traded Value of Puts to Calls 0.63 0.77 0.72 0.78 4.19 1.23 0.97
Odd-lots Volumes:
NYSE Odd Lot Purchases 0.00 −4.90 0.00 −5.89 −0.23 −0.16 0.20
NYSE Odd Lot Sales 0.11 0.27 0.30 0.59 −0.02 −0.02 0.06
NYSE Odd Lot Shorts 0.00 1.14 0.00 1.11 0.01 1.53 1.26
Short Sales Volumes:
NYSE Short Sales-Members 6.68 7.15 5.20 4.63 9.65 4.81 3.74
NYSE Short Sales-General public 2.63 2.58 1.53 1.67 7.84 2.64 4.13
NYSE Short Sales-Specialists 5.90 5.82 4.92 3.46 7.41 4.67 1.36
NYSE Short Sales-Total 6.80 5.59 5.13 4.22 11.57 4.94 5.94
Short Interests:
NYSE Short Interest Ratio −23.19 −2.22 −20.54 −2.01 −45.56 −1.48 0.60
NYSE Short Interest Shares −2.93 −0.12 −3.87 −0.16 −2.32 −0.04 0.00
AAII/II Sentiment Indices:
AAII Bearish Index 0.02 0.01 1.19 0.65 −3.43 −0.45 0.35
AAII Bullish Index 6.39 2.26 2.09 0.76 19.05 2.22 3.53
AAII Neutral Index −8.70 −2.83 −5.89 −1.85 −24.61 −2.79 4.03
Investors Intelligence Bearish Percentage −1.04 −0.11 7.12 0.75 −56.87 −1.27 1.96
Investors Intelligence Bullish Percentage −0.36 −0.03 −8.97 −0.92 28.17 0.74 0.89
Confidence Index:
Barron’s Confidence Index 36.44 0.78 −54.55 −1.13 182.58 1.82 4.54
Exchange Seat Prices:
AMEX Seat Prices 3.38 0.48 3.95 0.69 −0.25 −0.01 0.01
NYSE Annual Seat Price −16.55 −0.73 −31.32 −0.91 −33.93 −0.37 0.00
Volatility Indices:
CBOE S&P 500 Volatility Index 7.01 1.74 5.56 1.13 176.30 1.47 0.86
CBOE NASDAQ Volatility Index 13.28 2.10 17.01 3.21 15.40 0.53 0.10
CBOE S&P 100 Volatility Index 7.33 1.96 6.19 1.36 12.57 1.11 0.27
AMEX NYSE Arca NASDAQ 100 Volatility 4.00 0.61 10.92 2.67 −5.71 −0.38 1.11
Index
CBOE DJIA Volatility Index 13.39 1.90 9.81 2.24 22.21 0.97 0.28
Margin Account Balances:
NYSE Margin Debt −0.72 −0.02 −17.78 −0.42 35.82 0.41 0.30
NYSE Free Credit Balances 80.49 2.11 56.75 1.62 176.92 1.45 0.90
NYSE Free Credit Balances on Cash 22.34 0.63 32.78 0.96 −64.18 −0.42 0.39
Accounts
NYSE Free Cash Balances in Margin 1.66 0.04 −22.78 −0.70 69.74 0.81 1.02
Accounts
Mutual Fund Balances:
USA Mutual Fund Equity Funds Total Net 92.74 1.44 −6.10 −0.08 376.29 3.05 7.25
Assets
USA Mutual Fund Equity Funds Cash −20.76 −0.69 25.89 0.88 −288.05 −2.68 7.91
Percentage
USA Mutual Fund Equity Funds −4.74 −2.89 −4.53 −3.08 −2.93 −0.05 0.00
Redemptions
USA Mutual Fund Equity Funds New Sales 6.59 0.54 1.20 0.10 56.28 1.05 1.00

USA Mutual Fund Equity and Bond Fund 10.50 6.14 6.24 1.26 269.91 2.33 2.80
Net Assets
USA Mutual Fund Equity and Bond Fund −17.87 −0.78 10.37 0.47 −240.46 −2.57 6.80
Cash Percent
USA Mutual Fund Equity and Bond Fund 13.26 0.51 26.46 1.03 −72.82 −0.73 0.94
Liquid Assets
USA Mutual Fund Equity and Bond Fund −10.50 −0.91 −8.75 −0.76 −15.27 −0.37 0.02
Redemptions
USA Mutual Fund Equity and Bond Fund 7.89 0.85 3.31 0.37 27.69 0.97 0.66
New Sales
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 47

Table 4 (continued)

Market indicators Full sample Expansions Contractions Chi-statistic

β(∗10−3 ) t-stats β1 (∗10−3 ) t-stats β2 (∗10−3 ) t-stats

Number of dividend News:

Moody’s Monthly Decreased Dividends 40.61 1.57 66.23 2.80 4.40 0.07 0.79
Moody’s Monthly Extra Dividends −63.32 −1.38 −30.23 −0.68 −413.47 −2.01 3.31
Declared
Moody’s Monthly Increased Dividends −97.86 −1.97 −93.18 −1.85 −75.62 −0.47 0.01
Declared
Moody’s Monthly Omitted Dividends 7.60 0.24 39.20 1.26 −98.80 −1.14 2.23
Moody’s Monthly Resumed Dividends 15.28 0.81 32.23 1.93 10.13 0.13 0.08
S&P Monthly Dividend Decreases Declared 0.43 0.45 0.60 0.62 3.34 0.60 0.23
S&P Monthly Extra Dividends Declared 4.48 2.17 7.31 3.79 −12.79 −1.64 6.26
S&P Monthly Increased Dividends 2.11 0.57 7.93 1.43 −2.87 −0.75 2.63
Declared
S&P Monthly Omitted Dividends Declared 0.88 0.68 1.21 0.89 −2.23 −0.46 0.46
S&P Monthly Resumed Dividends 2.85 1.89 4.72 3.75 −3.18 −0.67 2.66
Declared
Panel B: Market Strength Indicators
Total Volume:
NYSE Total Volume 0.09 0.83 0.05 0.67 1.59 1.84 3.14
Total Volume Turnovers:
NYSE Share Volume Turnover 5.39 0.13 34.14 1.08 16.04 0.14 0.02
NYSE Annual Share Value Turnover 28.23 0.64 35.07 0.84 18.58 0.05 0.00
Short-term Trading Indices:
NYSE Short-term Trading Index −0.49 −2.15 −0.33 −1.25 −1.39 −2.12 2.30
NASDAQ Short-term Trading Index −0.01 −1.16 0.00 0.58 −0.02 −0.93 1.02
Daily Total Market Advances and Declines:
NYSE Advances 0.51 2.98 0.84 5.15 −0.15 −0.38 5.40
NYSE Declines −0.72 −3.65 −0.80 −4.53 −0.62 −1.31 0.12
NYSE Net Advances 0.00 0.49 0.00 0.22 0.00 0.24 0.02
NYSE AD Line 0.00 −0.35 0.00 −0.96 0.00 0.70 0.92
NYSE Percentage Net Advances 0.00 0.36 0.00 0.08 0.00 0.19 0.02
NASDAQ Advances 0.23 1.48 0.36 1.06 0.22 1.17 0.14
NASDAQ Declines −0.10 −3.41 −0.08 −5.74 −1.68 −1.45 1.90
NASDAQ Net Advances 0.00 −0.50 0.00 −0.59 0.02 1.05 1.29
NASDAQ AD Line 0.00 −0.22 0.00 0.36 −0.01 −0.76 0.70
NASDAQ Percentage Net Advances 0.00 −0.51 0.00 −0.60 0.02 1.05 1.29
Alternext Advances 1.18 4.02 1.18 3.22 1.20 2.42 0.00
Alternext Declines −1.04 −2.46 −0.80 −2.05 −2.66 −2.29 2.31
Alternext Net advances 0.01 0.80 0.00 0.07 0.02 0.75 0.52
Alternext AD Line 0.00 −0.03 −0.01 −0.96 0.01 0.47 0.49
Alternext Percentage Net Advances 0.01 0.60 0.00 −0.07 0.03 0.79 0.62
Weekly Total Market Advances and
Declines:
NYSE Weekly Advances −1.49 −3.33 −0.44 −0.77 −2.35 −4.08 5.61
NYSE Weekly Declines 0.65 1.21 0.71 1.31 0.79 0.55 0.00
NYSE Net Advances 0.00 0.22 0.00 0.27 −0.03 −0.19 0.04
NYSE AD Line −1.20 −0.52 −1.24 −0.53 1.54 0.02 0.00
Daily Total Market New Highs and New
Lows:
NYSE New Highs 0.14 3.61 0.52 3.89 0.11 7.73 9.38
NYSE New Lows −0.13 −1.50 −0.16 −1.58 0.01 0.07 0.63
NYSE Net New Highs 0.04 1.77 0.05 2.34 −0.06 −1.24 4.76
NYSE Cumulative Highs −0.01 −0.34 0.00 0.18 −0.02 −0.50 0.28
NYSE Percentage Net New Highs 0.04 1.60 0.05 2.08 −0.06 −1.30 4.55
NASDAQ New Highs −0.16 −0.43 0.24 0.73 −0.99 −1.03 1.47
NASDAQ New Lows 0.25 1.26 0.02 0.08 1.10 1.65 2.40
NASDAQ Net New Highs −0.01 −0.21 −0.01 −0.40 −0.01 −0.11 0.02
NASDAQ Cumulative Highs 0.03 0.98 0.03 0.89 0.04 0.26 0.01
NASDAQ Percentage Net New Highs −0.01 −0.22 −0.01 −0.41 −0.01 −0.12 0.02
Alternext New Highs 0.20 2.23 0.17 2.41 0.36 0.97 0.25
Alternext New Lows −0.06 −0.89 −0.16 −0.13 0.20 0.35 0.13
Alternext Net New Highs 0.00 0.11 −0.03 −0.59 0.07 1.16 1.60
Alternext Cumulative Highs −0.03 −0.88 −0.05 −1.16 −0.10 −1.32 0.42
Alternext Percentage Net New Highs 0.01 0.30 −0.02 −0.49 0.07 1.18 1.51

(continued on next page)

48 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Table 4 (continued)

Market indicators Full sample Expansions Contractions Chi-statistic

β(∗10−3 ) t-stats β1 (∗10−3 ) t-stats β2 (∗10−3 ) t-stats

Weekly Total Market New Highs and New

Lows:
NYSE Weekly New Highs 0.11 0.26 0.29 0.62 −0.21 −0.23 0.23
NYSE Weekly New Lows −0.30 −0.74 −0.45 −1.02 0.07 0.09 0.29
NYSE Net New Highs 0.11 1.88 0.08 1.43 0.31 2.24 2.32
NYSE Cumulative Highs −0.01 −3.62 −0.01 −7.07 −6.04 −4.57 20.82
This table reports the OLS results of the regression model Rt = αt + β1 Dt −1 It −1 + β2 (1 − Dt −1 )It −1 + εt . Rt represents S&P 500 periodic returns calculated
as log differences of the S&P 500 Index values, It −1 represents periodic percentage changes of market indicators. Dt −1 is a dummy variable that equals
1(0) during NBER business cycle expansions(contractions). Therefore β1 and β2 measure the predictability of a market indicator during expansions and
contractions respectively. We replicate the full sample OLS results for comparison in the first two columns, then we report β1 and β2 with associated
t-statistics, and the last column reports chi-statistics testing the null hypothesis that β1 and β2 are equal. We obtain all data from the Global Financial Data.
The t-statistics and chi-statistics reported are White standard errors corrected and marked in bold if significant at 10% significance level. Panel A and Panel
B report results for market sentiment and market strength indicators respectively.

periods, when sentiment becomes more important in driv- Moreover, we also use an alternative dataset to define
ing prices. Instead, we even have a few more predictive in- the business cycles—the CFNAI (Chicago Fed National Ac-
dicators during the low-sentiment period. Moreover, both tivity Index)16 data that starts from 1967. Compare with
numbers of significant predictors are less than the 30 found the NBER data, the CFNAI data is published in real time and
under the full-sample periods. Moreover, the F -test re- thus is free of the hindsight bias. The results stay similar,
sults also show that, statistically, 83 out of the 93 indica- no indicator predict market significantly under either con-
tors do not predict the market differently in two regimes. tractions or expansions. Last but not least, we also check if
This finding contributes to the view that separate high- and our results are sensitive to the 2008 financial crisis period,
low-sentiment regimes does not seem to increase the pre- also we remove top and bottom 5% extreme observations
dictability of the market indicators. from the distribution of each market indicator to control
To further check if any of the single-state predictive for outliers from the predictive variable direction. Our re-
indicators show true predictive value, we also perform sults stay robust. To save space, we present the detailed
similar economic significance tests as that above and results on these robustness checks in the appendix.
document the results in Tables 8 and 9 for the two regimes.
However, we find that none of the market indicators 7. Conclusion
outperform the market in terms of the Sharpe ratio or
Jensen’s α . The results show that predictability does not We review the predictability of a wide range of 93
strengthen under different sentiment regimes and our technical market indicators in predicting the S&P 500
main conclusion in the full-sample remains robust. returns. This adds to the literature with evidence from
widely used but less examined market indicators, to more
6. Robustness checks conclusively answer the question of whether technical
analysis is useful or not. Overall, we do not find the market
indicators generate profits that beat the buy and hold
We perform several robustness checks and find our
strategy. This result does not change if we consider the
conclusions hold. First, in the previous OLS rolling window
possibility of regime-switching predictability on business
regression analysis, quite a few of our market indicators
cycles or sentiment cycles. Moreover, our results remain
exhibit the widening of the confidence bounds problem.
robust if we use a GARCH(1, 1) or robust regression
And this may be a sign of volatility clustering. Therefore we
method. With previous mixed findings on price-based
used the GARCH(1, 1) instead of the OLS model to replicate
technical indicators, it is still not easy to provide a simple
our analysis. On the other hand, outliers can be another
positive or negative answer to the broad question of
issue that may cause the instability of indication. To deal
whether or not technical analysis is useful. Our results,
with this issue, we also replicate our analysis using robust
at least, make the answer not inconclusive with evidence
regressions to control the effect of potential outliers in the
from the family of market indicators missing.
dependent variable side. In addition, in previous analysis
we only test the economic significance of the market
indicators that show significant predictability in both sub- Appendix A. Supplementary data
samples under several alternative models. This may be
too restrictive. We loosen our criteria, we additionally test Supplementary material related to this article can be
the economic significance for the indicators that show found online at http://dx.doi.org/10.1016/j.jbef.2014.09.
significant predictability in the full-sample analysis but 001.
not in the sub-sample analysis. After all these checks, we
find technical market indicators still show very limited
predictive ability. 16 http://www.chicagofed.org/webpages/publications/cfnai/.
Table 5
NBER expansions economic significance test.
Market indicators Frequency N Buy and hold strategy Technical strategy
Mean Std. Dev. Sharpe ratio Mean Std. Dev. Sharpe ratio t-stats α (∗10−3 ) t-stats β t-stats
(∗10−3 ) (∗10−2 ) (∗10−2 ) (∗10−3 ) (∗10−2 ) (∗10−2 )
Panel A: Market Sentiment Indicators
NYSE Odd Lot Purchases Daily 5 202 0.25 1.17 1.30 0.31 0.85 2.49 1.16 0.13 1.61 0.54 22.54
NYSE Short Weekly 1 760 1.28 2.27 0.85 1.27 1.47 1.19 0.17 0.11 0.40 0.43 11.74
Sales-Members
NYSE Short Sales-General Weekly 1 738 1.28 2.27 0.85 1.61 1.63 3.21 1.34 0.43 1.55 0.52 13.56
Public
NYSE Short Weekly 1 758 1.28 2.27 0.85 1.49 1.45 2.74 0.95 0.33 1.23 0.42 11.44
Sales-Specialists
NYSE Short Sales-Total Weekly 1 756 1.28 2.27 0.85 1.46 1.50 2.45 0.83 0.30 1.09 0.45 12.11
NYSE Short Interest Ratio Monthly 458 5.29 4.51 1.34 6.55 3.48 5.34 1.27 1.17 1.10 0.60 10.96
AAII Neutral Index Weekly 563 −0.21 2.66 −2.43 0.23 1.33 −1.58 0.20 −0.06 −0.13 0.25 5.96
CBOE NASDAQ Volatility Daily 1 254 0.01 1.57 −0.29 −0.01 0.35 −1.93 0.47 −0.05 −0.57 0.05 5.62
Index
AMEX NYSE Arca NASDAQ Daily 1 253 0.01 1.57 −0.30 −0.05 0.35 −2.94 0.75 −0.09 −0.95 0.05 5.61
100 Volatility Index
CBOE DJIA Volatility Index Daily 732 −0.03 1.88 −0.20 0.01 0.00 0.00 0.04 0.00 N/A 0.00 N/A
USA Mutual Fund Equity Monthly 144 1.29 4.87 −3.16 4.91 3.94 5.27 1.71 2.06 1.12 0.69 7.28
Funds Redemptions
Moody’s Monthly Monthly 310 6.16 4.57 5.76 5.24 3.56 4.81 0.25 −0.53 −0.39 0.61 9.02
Decreased Dividends
Moody’s Monthly Monthly 310 6.16 4.57 5.76 5.65 3.61 5.89 0.04 −0.16 −0.12 0.63 9.26
Increased Dividends
Declared
Moody’s Monthly Monthly 310 6.16 4.57 5.76 5.66 3.46 6.16 0.10 1.11 0.81 0.52 6.73
Resumed Dividends
S&P Monthly Extra Monthly 317 6.36 4.54 5.03 6.38 2.99 7.71 0.58 0.82 0.61 0.44 6.37
Dividends Declared
S&P Monthly Resumed Monthly 312 7.30 4.43 7.01 7.23 3.28 9.25 0.55 0.01 0.01 0.58 8.33
Dividends Declared
Panel B: Market Strength Indicators
NYSE Advances Daily 10 975 0.24 1.06 0.86 0.00 0.67 −2.24 3.71 −0.19 −3.88 0.39 17.56
NYSE Declines Daily 10 975 0.24 1.06 0.86 0.09 0.76 −0.83 2.33 −0.12 −2.28 0.51 19.56
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

NASDAQ Declines Daily 4 955 0.25 1.18 1.33 0.30 0.87 2.44 1.08 0.13 1.50 0.55 22.04
Alternext Advances Daily 6 573 0.30 1.18 1.59 −0.08 0.47 −4.15 4.13 −0.22 −4.27 0.14 8.22
Alternext Declines Daily 6 573 0.30 1.18 1.59 −0.28 0.62 −6.34 6.49 −0.46 −6.90 0.27 11.51
NYSE New Highs Daily 10 260 0.26 1.07 1.07 0.14 0.14 −0.83 1.39 −0.02 −2.22 0.00 5.08
NYSE Net New Highs Daily 10 093 0.31 1.08 1.51 0.35 0.83 2.39 1.29 0.10 1.84 0.60 26.43
NYSE Percentage Net New Daily 10 091 0.26 1.08 1.01 0.30 0.82 1.86 1.24 0.09 1.64 0.59 23.05
Highs
Alternext New Highs Daily 6 076 0.26 1.20 1.25 0.27 0.94 1.74 0.58 0.07 0.88 0.62 19.22
NYSE Cumulative Highs Weekly 1 937 1.64 2.32 2.63 1.74 1.63 4.34 0.98 0.40 1.52 0.49 15.56
This table compares risk and return characteristics of buy and hold strategies with technical strategies based on market indicators that show significant predictability during NBER expansion periods. We report
means, standard deviations and Sharpe ratios for buy and hold strategies and technical strategies consequently, then we report t-statistics testing the null hypothesis that the Sharpe ratios of the two strategies
are equal. We then report Jensen’s α estimation results for technical strategies in the last four columns. α values indicate excess returns generated by the technical strategies at given risk level β over market
returns. We also present associated t-statistics testing their differences from zero for α and β values. All t-statistics are White standard errors corrected and marked in bold if significant at 10% significance level.
The estimations are based on the OLS regression results.
49
 50

Table 6
NBER contractions economic significance test.
Market indicators Frequency N Buy and hold strategy Technical strategy
Mean Std. Dev. Sharpe Mean Std. Dev. Sharpe t- α (∗10−3 ) t-stats β t-stats
(∗10−3 ) (∗10−2 ) ratio (∗10−3 ) (∗10−2 ) ratio stats
(∗10−2 ) (∗10−2 )
Panel A: Market Sentiment Indicators
NYSE Short Sales-Members Weekly 1 780 1.26 2.26 0.76 1.18 0.82 1.10 0.13 0.07 0.39 0.14 2.98
NYSE Short Sales-General Public Weekly 1 783 1.26 2.26 0.76 1.11 0.66 0.26 0.18 0.00 0.03 0.09 2.01
NYSE Short Sales-Specialists Weekly 1 781 1.26 2.26 0.76 1.09 0.77 0.01 0.28 −0.02 −0.10 0.12 2.72
NYSE Short Sales-Total Weekly 1 782 1.26 2.26 0.76 1.12 0.77 0.39 0.14 0.01 0.07 0.12 2.69
AAII Bullish Index Weekly 563 −0.21 2.66 −2.43 0.36 0.87 −0.86 0.32 −0.01 −0.04 0.11 2.54
AAII Neutral Index Weekly 563 −0.21 2.66 −2.43 0.56 1.22 1.02 0.79 0.25 0.57 0.21 2.44
Barron’s Confidence Index Weekly 2 063 1.24 2.31 0.92 0.99 0.54 −0.81 0.63 −0.06 −0.49 0.06 3.69
USA Mutual Fund Equity Funds Total Net Monthly 321 6.46 4.54 5.22 3.39 1.47 −4.74 1.42 −0.03 −0.06 0.05 2.20
Assets
USA Mutual Fund Equity Funds Cash Monthly 321 6.46 4.54 5.22 3.32 1.96 −3.94 1.44 −0.37 −0.39 0.14 2.33
Percentage
USA Mutual Fund Equity and Bond Fund Monthly 147 1.75 4.89 −2.22 1.41 1.92 −7.44 0.51 0.14 0.14 0.07 1.82
Net Assets
USA Mutual Fund Equity and Bond Fund Monthly 241 4.92 4.36 3.44 2.92 1.59 −3.18 0.83 0.26 0.38 0.06 2.21
Cash Percent
Moody’s Monthly Extra Dividends Declared Monthly 310 6.16 4.57 5.76 2.06 1.74 −8.42 2.10 −1.12 −1.22 0.10 1.72
Panel B: Market Strength Indicators
NYSE Total Volume Daily 10 844 0.24 1.06 0.84 0.15 0.17 0.00 0.64 −0.01 −0.85 0.01 2.63
NYSE Short-term Trading Index Daily 5 725 0.28 1.14 1.51 −0.02 0.47 −2.70 2.88 −0.16 −2.90 0.15 7.97
Alternext Advances Daily 6 472 0.30 1.18 1.59 0.01 0.32 −3.22 3.13 −0.12 −3.40 0.06 6.07
Alternext Declines Daily 6 472 0.30 1.18 1.59 −0.01 0.43 −2.94 3.16 −0.15 −3.15 0.11 7.32
NYSE Weekly Advances Weekly 1 841 1.47 2.26 2.01 0.93 0.94 −0.94 1.17 −0.17 −0.83 0.17 3.98
NYSE New Highs Daily 10 130 0.26 1.07 1.07 0.16 0.16 0.77 0.22 0.00 0.42 0.01 3.28
NASDAQ New Lows Daily 4 534 0.24 1.21 1.26 0.02 0.50 −1.32 1.57 −0.10 −1.50 0.15 7.48
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

NYSE Net New Highs Weekly 1 927 1.61 2.32 2.52 0.90 0.51 −2.52 1.77 −0.16 −1.22 0.05 1.48
NYSE Cumulative Highs Weekly 1 937 1.64 2.32 2.63 0.88 1.23 −1.26 1.77 −0.33 −1.36 0.28 7.48
This table compares risk and return characteristics of buy and hold strategies with technical strategies based on market indicators that show significant predictability during NBER contraction periods. We report
means, standard deviations and Sharpe ratios for buy and hold strategies and technical strategies consequently, then we report t-statistics testing the null hypothesis that the Sharpe ratios of the two strategies
are equal. We then report Jensen’s α estimation results for technical strategies in the last four columns. α values indicate excess returns generated by the technical strategies at given risk level β over market
returns. We also present associated t-statistics testing their differences from zero for α and β values. All t-statistics are White standard errors corrected and marked in bold if significant at 10% significance level.
The estimations are based on the OLS regression results.
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 51

Table 7
Sentiment cycle time-varying results.
Market indicators Full sample High sentiment Low sentiment Chi-statistic
β (∗10−3 ) t-stats β1 (∗10−3 ) t-stats β2 (∗10−3 ) t-stats

Panel A: Market Sentiment Indicators

Option Volumes:
CBOE Calls Volume 0.00 1.15 −0.19 −0.45 −0.65 −1.10 0.40
CBOE Puts Volume −0.01 −0.12 0.01 0.38 −0.24 −0.48 0.26
OEX Calls Volume 0.00 −1.26 0.00 −1.21 −0.31 −1.40 1.95
OEX Puts Volume 0.00 0.25 0.00 0.25 0.18 0.55 0.30
CBOE Ratio of Traded Value of Puts to Calls 0.63 0.77 0.59 0.50 0.82 0.74 0.02
Odd-lots Volumes:
NYSE Odd Lot Purchases 0.00 −4.90 −0.14 −0.21 0.00 −6.68 0.04
NYSE Odd Lot Sales 0.11 0.27 0.61 1.17 −0.16 −0.24 0.87
NYSE Odd Lot Shorts 0.00 1.14 0.01 1.46 0.00 0.29 1.61
Short Sales Volumes:
NYSE Short Sales-Members 6.68 7.15 7.39 4.92 7.63 4.43 0.01
NYSE Short Sales-General Public 2.63 2.58 6.79 2.25 4.70 2.16 0.32
NYSE Short Sales-Specialists 5.90 5.82 5.93 3.63 6.51 3.64 0.06
NYSE Short Sales-Total 6.80 5.59 8.62 3.62 8.45 4.30 0.00
Short Interests:
NYSE Short Interest Ratio −23.19 −2.22 −9.58 −0.32 67.65 2.19 3.33
NYSE Short Interest Shares −2.93 −0.12 43.55 0.86 11.86 0.29 0.25
AAII/II Sentiment Indices:
AAII Bearish Index 0.02 0.01 1.82 0.68 −2.05 −0.75 1.05
AAII Bullish Index 6.39 2.26 4.98 1.32 7.87 1.92 0.27
AAII Neutral Index −8.70 −2.83 −10.55 −2.44 −6.71 −1.55 0.40
Investors Intelligence Bearish Percentage −1.04 −0.11 14.85 1.12 −18.18 −1.39 3.14
Investors Intelligence Bullish Percentage −0.36 −0.03 −7.94 −0.56 5.39 0.34 0.39
Confidence Index:
Barron’s Confidence Index 36.44 0.78 −95.13 −1.20 38.75 0.46 1.33
Exchange Seat Prices:
AMEX Seat Prices 3.38 0.48 4.63 0.28 −2.62 −0.63 0.19
NYSE Annual Seat Price −16.55 −0.73 31.87 0.30 65.06 0.55 0.05
Volatility Indices:
CBOE S&P 500 Volatility Index 7.01 1.74 4.71 0.77 12.19 2.44 0.90
CBOE NASDAQ Volatility Index 13.28 2.10 3.00 0.46 18.48 1.99 1.86
CBOE S&P 100 Volatility Index 7.33 1.96 5.61 1.00 10.80 2.30 0.50
AMEX NYSE Arca NASDAQ 100 Volatility Index 4.00 0.61 3.22 0.61 3.96 0.44 0.00
CBOE DJIA Volatility Index 13.39 1.90 9.06 1.93 15.91 1.37 0.30
Margin Account Balances:
NYSE Margin Debt −0.72 −0.02 −55.63 −0.83 35.59 0.47 0.85
NYSE Free Credit Balances 80.49 2.11 2.42 0.04 67.13 1.08 0.60
NYSE Free Credit Balances on Cash Accounts 22.34 0.63 58.72 1.11 −7.45 −0.16 0.87
NYSE Free Cash Balances in Margin Accounts 1.66 0.04 −38.28 −1.03 50.08 0.95 1.91
Mutual Fund Balances:
USA Mutual Fund Equity Funds Total Net Assets 92.74 1.44 49.93 0.65 142.33 1.50 0.66
USA Mutual Fund Equity Funds Cash Percentage −20.76 −0.69 −74.82 −1.65 21.77 0.56 2.60
USA Mutual Fund Equity Funds Redemptions −4.74 −2.89 −5.32 −4.64 10.52 0.35 0.28
USA Mutual Fund Equity Funds New Sales 6.59 0.54 2.44 0.18 16.16 0.60 0.21
USA Mutual Fund Equity and Bond Fund Net 10.50 6.14 127.09 2.09 9.66 11.91 3.74
Assets
USA Mutual Fund Equity and Bond Fund Cash −17.87 −0.78 −61.17 −1.42 6.46 0.20 1.59
Percent
USA Mutual Fund Equity and Bond Fund Liquid 13.26 0.51 −7.66 −0.19 −190.00 0.63 0.34
Assets
USA Mutual Fund Equity and Bond Fund −10.50 −0.91 −18.90 −0.99 −5.32 −0.28 0.25
Redemptions
USA Mutual Fund Equity and Bond Fund New 7.89 0.85 15.15 0.99 5.21 0.39 0.24
Sales
Number of Dividend News:
Moody’s Monthly Decreased Dividends 40.61 1.57 40.15 0.92 28.67 0.63 0.03
Moody’s Monthly Extra Dividends Declared −63.32 −1.38 −49.04 −0.81 −72.71 −0.96 0.06
Moody’s Monthly Increased Dividends Declared −97.86 −1.97 −29.62 −0.32 −159.24 −1.93 1.07
Moody’s Monthly Omitted Dividends 7.60 0.24 23.30 0.46 −19.25 −0.36 0.32
Moody’s Monthly Resumed Dividends 15.28 0.81 39.00 2.02 −14.91 −0.46 2.05

(continued on next page)

52 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Table 7 (continued)

Market indicators Full sample High sentiment Low sentiment Chi-statistic

β (∗10−3 ) t-stats β1 (∗10−3 ) t-stats β2 (∗10−3 ) t-stats

S&P Monthly Dividend Decreases Declared 0.43 0.45 0.34 0.13 0.58 0.72 0.01
S&P Monthly Extra Dividends Declared 4.48 2.17 5.04 1.46 3.07 0.91 0.17
S&P Monthly Increased Dividends Declared 2.11 0.57 −0.93 −0.26 7.81 1.08 1.19
S&P Monthly Omitted Dividends Declared 0.88 0.68 1.10 0.73 −0.07 −0.03 0.19
S&P Monthly Resumed Dividends Declared 2.85 1.89 7.01 3.16 0.71 0.35 4.72

Panel B: Market Strength Indicators

Total Volume:
NYSE Total Volume 0.09 0.83 0.61 2.24 0.54 1.74 0.03
Total Volume Turnovers:
NYSE Share Volume Turnover 5.39 0.13 23.08 0.57 −113.88 −1.69 3.05
NYSE Annual Share Value Turnover 28.23 0.64 194.47 1.43 −207.02 −1.47 4.28
Short-term Trading Indices:
NYSE Short-term Trading Index −0.49 −2.15 −0.49 −1.11 −0.49 −2.01 0.00
NASDAQ Short-term Trading Index −0.01 −1.16 0.00 0.85 −0.02 −1.28 2.02
Daily Total Market Advances and Declines:
NYSE Advances 0.51 2.98 1.18 2.39 0.66 1.88 0.74
NYSE Declines −0.72 −3.65 −1.13 −3.47 −1.31 −2.57 0.08
NYSE Net Advances 0.00 0.49 0.00 0.79 0.00 0.75 0.11
NYSE AD Line 0.00 −0.35 0.00 0.19 −0.01 −0.71 0.53
NYSE Percentage Net Advances 0.00 0.36 0.00 0.79 0.00 0.76 0.11
NASDAQ Advances 0.23 1.48 −0.13 −0.28 0.28 1.65 0.65
NASDAQ Declines −0.10 −3.41 −0.07 −11.09 −1.27 −2.46 5.35
NASDAQ Net Advances 0.00 −0.50 0.00 0.43 0.00 −0.94 0.54
NASDAQ AD Line 0.00 −0.22 0.00 0.44 0.00 −0.46 0.41
NASDAQ Percentage Net Advances 0.00 −0.51 0.00 0.41 0.00 −0.94 0.51
Alternext Advances 1.18 4.02 1.32 1.87 1.01 3.54 0.16
Alternext Declines −1.04 −2.46 −1.16 −2.06 −0.87 −1.95 0.16
Alternext Net Advances 0.01 0.80 −0.01 −0.35 0.01 1.04 0.80
Alternext AD Line 0.00 −0.03 0.00 0.31 0.00 0.08 0.04
Alternext Percentage Net Advances 0.01 0.60 0.00 0.00 0.01 1.06 0.83
Weekly Total Market Advances and Declines:
NYSE Weekly Advances −1.49 −3.33 0.47 0.47 −2.24 −5.14 6.16
NYSE Weekly Declines 0.65 1.21 1.64 1.46 0.41 0.40 0.64
NYSE Net Advances 0.00 0.22 0.01 0.75 −0.01 −0.36 0.58
NYSE AD Line −1.20 −0.52 −3.08 −0.03 −21.32 −0.38 0.03
Daily Total Market New Highs and New Lows:
NYSE New Highs 0.14 3.61 0.13 0.66 0.11 8.51 0.01
NYSE New Lows −0.13 −1.50 −0.33 −1.19 −0.18 −1.01 0.22
NYSE Net New Highs 0.04 1.77 0.03 1.25 0.00 0.10 0.40
NYSE Cumulative Highs −0.01 −0.34 −0.02 −0.50 0.00 0.13 0.24
NYSE Percentage Net New Highs 0.04 1.60 0.03 1.35 0.00 0.16 0.42
NASDAQ New Highs −0.16 −0.43 −0.17 −0.43 −0.13 −0.22 0.00
NASDAQ New Lows 0.25 1.26 0.31 0.79 0.21 0.96 0.04
NASDAQ Net New Highs −0.01 −0.21 −0.03 −0.71 0.04 0.70 0.96
NASDAQ Cumulative Highs 0.03 0.98 0.06 1.91 −0.01 −0.25 1.81
NASDAQ Percentage Net New Highs −0.01 −0.22 −0.03 −0.76 0.04 0.75 1.09
Alternext New Highs 0.20 2.23 0.60 2.39 0.29 1.82 1.07
Alternext New Lows −0.06 −0.89 −0.03 −0.14 −0.07 −0.82 0.03
Alternext Net New Highs 0.00 0.11 −0.04 −0.64 0.04 0.71 0.89
Alternext Cumulative Highs −0.03 −0.88 −0.08 −1.59 0.00 0.04 1.16
Alternext Percentage Net New Highs 0.01 0.30 −0.03 −0.58 0.04 0.81 0.95
Weekly Total Market New Highs and New Lows:
NYSE Weekly New Highs 0.11 0.26 1.42 2.26 −0.24 −0.37 3.43
NYSE Weekly New Lows −0.30 −0.74 0.78 0.52 −0.77 −0.79 0.76
NYSE Net New Highs 0.11 1.88 0.66 2.13 0.03 0.56 4.01
NYSE Cumulative Highs −0.01 −3.62 31.60 0.25 −0.01 −3.46 0.06
This table reports the OLS results of the regression model Rt = αt + β1 Dt −1 It −1 + β2 (1 − Dt −1 )It −1 + εt . Rt represents S&P 500 periodic returns calculated
as log differences of the S&P 500 Index values, It −1 represents periodic percentage changes of market indicators. Dt −1 is a dummy variable that equals 1(0)
during high (low) sentiment periods measured by using the Baker and Wurgler (2006) sentiment index. Therefore β1 and β2 measure the predictability of
a market indicator during expansions and contractions respectively. We replicate the full sample OLS results for comparison in the first two columns, then
we report β1 and β2 with associated t-statistics, and the last column reports chi-statistics testing the null hypothesis that β1 and β2 are equal. We obtain
all data from the Global Financial Data. The t-statistics and chi-statistics reported are White standard errors corrected and marked in bold if significant at
10% significance level. Panel A and Panel B report results for market sentiment and market strength indicators respectively.
Table 8
High sentiment periods economic significance test.
Market indicators Frequency N Buy and hold strategy Technical strategy
Mean Std. Dev. Sharpe ratio Mean Std. Dev. Sharpe ratio t- α (∗10−3 ) t-stats β t-
(∗10−3 ) (∗10−2 ) (∗10−2 ) (∗10−3 ) (∗10−2 ) (∗10−2 ) stats stats

Panel A: Market Sentiment Indicators

NYSE Short Sales-Members Weekly 930 0.86 2.36 0.13 0.13 1.03 −6.84 2.23 −0.62 −2.02 0.20 5.07
NYSE Short Sales-General Public Weekly 970 0.86 2.36 0.13 0.53 1.26 −2.36 0.87 −0.25 −0.70 0.28 7.01
NYSE Short Sales-Specialists Weekly 927 0.86 2.36 0.13 0.58 1.00 −2.45 0.81 −0.21 −0.69 0.18 5.09
NYSE Short Sales-Total Weekly 938 0.86 2.36 0.13 0.24 1.13 −5.19 1.77 −0.51 −1.53 0.23 5.84
AAII Neutral Index Weekly 508 −0.14 2.70 −2.19 0.33 1.02 −1.21 0.21 0.02 0.05 0.14 3.19
CBOE DJIA Volatility Index Daily 714 −0.11 1.92 −0.66 0.01 0.00 0.00 0.13 0.00 N/A 0.00 N/A
USA Mutual Fund Equity Funds Monthly 228 4.98 4.40 4.22 5.47 2.55 9.19 0.84 1.55 1.11 0.34 5.90
Cash Percentage
USA Mutual Fund Equity Funds Monthly 145 2.09 4.86 −0.85 2.63 3.29 0.38 0.19 −0.36 −0.18 0.47 5.41
Redemptions
USA Mutual Fund Equity and Bond Monthly 241 4.78 4.62 3.07 3.80 2.33 1.87 0.20 −0.11 −0.08 0.26 4.04
Fund Net Assets
Moody’s Monthly Resumed Monthly 243 5.77 4.35 5.62 4.46 2.61 4.34 0.23 −0.05 −0.04 0.35 5.53
Dividends
S&P Monthly Resumed Dividends Monthly 232 6.26 4.51 6.40 4.37 3.09 3.23 0.63 −0.13 −0.08 0.46 5.14
Declared
Panel B: Market Strength Indicators
NYSE Total Volume Daily 5585 0.27 1.14 1.44 0.19 0.70 1.25 0.16 0.04 0.49 0.37 19.24
NYSE Advances Daily 4843 0.27 1.14 1.44 −0.01 0.36 −3.21 2.99 −0.11 −2.23 0.10 11.35
NYSE Declines Daily 4987 0.27 1.14 1.44 −0.01 0.51 −2.22 2.61 −0.12 −1.88 0.19 13.10
NASDAQ Declines Daily 4910 0.24 1.18 1.27 0.25 0.75 2.06 0.65 0.09 1.15 0.41 19.02
Alternext Advances Daily 4902 0.27 1.14 1.44 −0.04 0.42 −3.50 3.30 −0.14 −2.58 0.13 11.33
Alternext Declines Daily 5056 0.27 1.14 1.44 0.01 0.53 −1.83 2.37 −0.11 −1.68 0.21 13.89
NASDAQ Cumulative Highs Daily 4541 0.23 1.21 1.20 0.26 0.78 2.14 0.75 0.11 1.21 0.42 18.75
Alternext New Highs Daily 4875 0.27 1.15 1.47 0.02 0.49 −1.74 2.21 −0.09 −1.37 0.18 14.00
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

NYSE Weekly New Highs Weekly 1199 1.36 2.34 2.61 0.70 1.27 −0.33 0.32 −0.17 −0.52 0.29 8.08
NYSE Net New Highs Weekly 1151 1.62 2.33 3.78 1.12 1.40 2.72 0.41 0.13 0.40 0.36 9.97
This table compares risk and return characteristics of buy and hold strategies with technical strategies based on market indicators that show significant predictability during high sentiment periods. We report
means, standard deviations and Sharpe ratios for buy and hold strategies and technical strategies consequently, then we report t-statistics testing the null hypothesis that the Sharpe ratios of the two strategies
are equal. We then report Jensen’s α estimation results for technical strategies in the last four columns. α values indicate excess returns generated by the technical strategies at given risk level β over market
returns. We also present associated t-statistics testing their differences from zero for α and β values. All t-statistics are White standard errors corrected and marked in bold if significant at 10% significance level.
The estimations are based on the OLS regression results.
53
 54

Table 9
Low sentiment periods economic significance test.
Market indicators Frequency N Buy and hold strategy Technical strategy
Mean Std. Dev. Sharpe ratio Mean Std. Dev. Sharpe ratio t- α (∗10−3 ) t-stats β t-
(∗10−3 ) (∗10−2 ) (∗10−2 ) (∗10−3 ) (∗10−2 ) (∗10−2 ) stats stats

Panel A: Market Sentiment Indicators

NYSE Odd Lot Purchases Daily 5175 0.24 1.17 1.26 0.01 0.89 −0.96 2.22 −0.17 −2.08 0.56 25.61
NYSE Short Sales-Members Weekly 1002 0.86 2.36 0.13 0.81 0.68 −0.31 0.13 −0.07 −0.34 0.09 3.32
NYSE Short Sales-General Public Weekly 1012 0.86 2.36 0.13 0.83 0.78 −0.05 0.05 −0.05 −0.19 0.12 4.11
NYSE Short Sales-Specialists Weekly 1005 0.86 2.36 0.13 0.76 0.72 −1.00 0.33 −0.14 −0.62 0.10 3.49
NYSE Short Sales-Total Weekly 1009 0.86 2.36 0.13 0.64 0.72 −2.57 0.79 −0.23 −1.01 0.10 3.66
NYSE Short Interest Ratio Weekly 263 4.78 4.62 3.07 2.64 2.74 −2.63 0.96 −0.63 −0.46 0.29 3.84
AAII Bullish Index Weekly 538 −0.14 2.70 −2.19 0.29 1.28 −1.30 0.20 −0.08 −0.16 0.22 4.70
CBOE S&P 500 Volatility Index Daily 2935 0.04 1.36 −0.25 0.05 0.83 −0.26 0.00 −0.08 −0.71 0.39 14.17
CBOE NASDAQ Volatility Index Daily 1069 −0.01 1.58 −0.45 0.20 0.76 1.86 0.79 0.00 −0.02 0.23 4.98
CBOE S&P 100 Volatility Index Daily 2956 0.04 1.36 −0.25 0.14 0.84 0.77 0.66 0.00 0.01 0.39 14.23
USA Mutual Fund Equity and Bond Monthly 266 4.78 4.62 3.07 2.23 2.71 −4.18 1.23 −1.03 −0.75 0.30 4.06
Fund Net Assets
Moody’s Monthly Increased Monthly 246 5.77 4.35 5.62 3.90 2.61 2.16 0.59 −0.05 −0.04 0.32 4.31
Dividends Declared
Panel B: Market Strength Indicators
NYSE Total Volume Daily 5512 0.27 1.14 1.44 0.08 0.81 −0.27 1.66 −0.10 −1.27 0.49 22.07
NYSE Share Volume Turnover Daily 258 4.78 4.62 3.07 4.04 2.34 2.93 0.02 1.00 0.84 0.21 4.08
NYSE Short-term Trading Index Daily 4959 0.27 1.14 1.41 −0.13 0.53 −4.38 4.22 −0.27 −4.13 0.22 10.60
NYSE Advances Daily 4901 0.27 1.14 1.44 −0.11 0.37 −5.77 4.69 −0.24 −4.76 0.11 6.67
NYSE Declines Daily 4918 0.27 1.14 1.44 −0.09 0.48 −4.01 3.83 −0.23 −3.77 0.18 10.07
NASDAQ Advances Daily 4403 0.24 1.18 1.27 −0.12 0.52 −4.01 3.51 −0.23 −3.35 0.20 10.82
NASDAQ Declines Daily 4273 0.24 1.18 1.27 −0.16 0.50 −4.92 4.04 −0.28 −4.02 0.18 9.60
Alternext Advances Daily 4931 0.27 1.14 1.44 −0.06 0.41 −4.07 3.66 −0.19 −3.62 0.13 8.57
Alternext Declines Daily 5100 0.27 1.14 1.44 0.07 0.66 −0.54 1.60 −0.13 −1.72 0.32 12.71
NYSE Weekly Advances Weekly 1059 1.36 2.34 2.61 0.63 1.36 −0.86 1.32 −0.22 −0.63 0.34 6.00
J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

NYSE New Highs Daily 5556 0.28 1.15 1.51 0.06 0.85 −0.59 2.16 −0.11 −1.47 0.54 24.73
Alternext New Highs Daily 5249 0.27 1.15 1.47 −0.01 0.72 −1.59 2.66 −0.16 −2.13 0.40 18.38
NYSE Cumulative Highs Weekly 1196 1.64 2.33 3.86 0.94 1.62 1.19 1.19 −0.27 −0.80 0.48 11.25
This table compares risk and return characteristics of buy and hold strategies with technical strategies based on market indicators that show significant predictability during low sentiment periods. We report
means, standard deviations and Sharpe ratios for buy and hold strategies and technical strategies consequently, then we report t-statistics testing the null hypothesis that the Sharpe ratios of the two strategies
are equal. We then report Jensen’s α estimation results for technical strategies in the last four columns. α values indicate excess returns generated by the technical strategies at given risk level β over market
returns. We also present associated t-statistics testing their differences from zero for α and β values. All t-statistics are White standard errors corrected and marked in bold if significant at 10% significance level.
The estimations are based on the OLS regression results.
 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56 55

References Franklin Fant, L., 1999. Investment Behavior of Mutual Fund Sharehold-
ers: The Evidence from Aggregate Fund Flows.
Abraham, A., Ikenberry, D.L., 1994. The individual investor and the Frazzini, A., Lamont, O.A., 2008. Dumb money: Mutual fund flows and the
weekend effect. J. Finan. Quant. Anal. 29 (02), 263–277. cross-section of stock returns. J. Financ. Econ. 88 (2), 299–322.
Achelis, S.B., 2001. Technical Analysis from A to Z. McGraw Hill, New York. Friesen, G.C., Sapp, T.R., 2007. Mutual fund flows and investor returns: An
Baker, M., Stein, J.C., 2004. Market liquidity as a sentiment indicator. J. empirical examination of fund investor timing ability. J. Bank. Finance
Financ. Mark. 7 (3), 271–299. 31 (9), 2796–2816.
Baker, M., Wurgler, J., 2006. Investor sentiment and the cross-section of Gaumnitz, J.E., Salabar, C.A., 1969. The Barron’s confidence index: An
stock returns. J. Finance 61 (4), 1645–1680. examination of its value as a market indicator. Financ. Anal. J. 25 (5),
Baker, M., Wurgler, J., 2007. Investor sentiment in the stock market. J. 16–70.
Econ. Perspect. 21 (2), 129–152. Giot, P., 2003. The information content of implied volatility in agricultural
Benartzi, S., Michaely, R., Thaler, R., 1997. Do changes in dividends signal commodity markets. J. Futures Mark. 23 (5), 441–454.
the future or the past? J. Finance 52 (3), 1007–1034. Glushkov, D., 2006. Sentiment beta. Available at SSRN: http://ssrn.com/
Ben-Rephael, A., Kandel, S., Wohl, A., 2012. Measuring investor sentiment abstract=862444.
with mutual fund flows. J. Financ. Econ. 104 (2), 363–382. Griffin, J.M., Ji, X., Martin, J.S., 2003. Momentum investing and business
Biggs, B.M., 1966. The short interest: A false proverb. Financ. Anal. J. 22 cycle risk: Evidence from pole to pole. J. Finance 58 (6), 2515–2547.
(4), 111–116. Gup, B.E., 1973. A note on stock market indicators and stock prices. J.
Blume, L., Easley, D., O’Hara, M., 1994. Market statistics and technical Finan. Quant. Anal. 8 (04), 673–682.
analysis: The role of volume. J. Finance 49 (1), 153–181. Han, B., 2008. Investor Sentiment and Option Prices.
Bowlin, L., Rozeff, M.S., 1987. Do Specialists’ Short Sales Predict Returns?. Han, Y., Yang, K., Zhou, G., 2013. A new anomaly: The cross-sectional
Boyd, J.H., Hu, J., Jagannathan, R., 2005. The stock market’s reaction to profitability of technical analysis. J. Finan. Quant. Anal. 48 (5),
unemployment news: Why bad news is usually good for stocks. J. 1433–1461.
Finance 60 (2), 649–672. Hanna, M., 1976. A stock price predictive model based on changes in
Branch, B., 1976. The predictive power of stock market indicators. J. Finan. ratios of short interest to trading volume. J. Finan. Quant. Anal. 11
Quant. Anal. 11 (2), 269–285. (5), 857–872.
Brock, W., Lakonishok, J., LeBaron, B., 1992. Simple technical trading Hendricks, D., Patel, J., Zeckhauser, R., 1993. Hot hands in mutual
rules and the stochastic properties of stock returns. J. Finance 47 (5), funds: Short—run persistence of relative performance, 1974–1988. J.
1731–1764. Finance 48 (1), 93–130.
Brown, G.W., Cliff, M.T., 2004. Investor sentiment and the near-term stock Henkel, S.J., Martin, J.S., Nardari, F., 2011. Time-varying short-horizon
market. J. Empir. Finance 11 (1), 1–27. predictability. J. Financ. Econ. 99 (3), 560–580.
Brown, G.W., Cliff, M.T., 2005. Investor sentiment and asset valuation. J. Hirose, T., Kato, H.K., Bremer, M., 2009. Can margin traders predict future
Bus. 78 (2), 405–440. stock returns in Japan? Pacific-Basin Financ. J. 17 (1), 41–57.
Chen, S., Gu, A.Y., 2009. Margin debt balance vs stock market movements Jacobsen, B., Dannenburg, D., 2003. Volatility clustering in monthly stock
and expected GDP growth. Acad. Account. Financ. Stud. 14 (2), 23. returns. J. Empir. Finance 10, 479–503.
Chiang, R., Gay, G., Kolb, R., 1987. Commodity exchange seat prices. Rev. Jacobsen, B., Marshall, B., Visaltanachoti, N., 2007. Return predictability
Futures Mark. 6, 1–10. revisited. Available from SSRN http://papers.ssrn.com/sol3/papers.
Chordia, T., Shivakumar, L., 2002. Momentum, business cycle, and time- cfm?abstract_id=965336.
varying expected returns. J. Finance 57 (2), 985–1019. Jensen, M.C., Benington, G.A., 1970. Random walks and technical theories:
Clarke, R.G., Statman, M., 1998. Bullish or bearish? Financ. Anal. J. 54 (3), Some additional evidence. J. Finance 25 (2), 469–482.
63–72. Kaish, S., 1969. Odd lot profit and loss performance. Financ. Anal. J. 25 (5),
Dangl, T., Halling, M., 2012. Predictive regressions with time-varying 83–89.
coefficients. J. Financ. Econ. 106 (1), 157–181. Keim, D.B., Madhavan, A., 2000. The relation between stock market
De Bondt, W.P., 1993. Betting on trends: Intuitive forecasts of financial movements and NYSE seat prices. J. Finance 55 (6), 2817–2840.
risk and return. Int. J. Forecast. 9 (3), 355–371. Kerrigan, T.J., 1974. The short interest ratio and its component parts.
De Long, J., Shleifer, A., Summers, L.H., Waldmann, R.J., 1989. The size and Financ. Anal. J. 30 (6), 45–49.
incidence of the losses from noise trading. J. Finance 44 (3), 681–696. Kewley, T.J., Stevenson, R.A., 1967. The odd-lot theory as revealed by
De Long, J.B., Shleifer, A., Summers, L.H., Waldmann, R.J., 1990. purchase and sale statistics for individual stocks. Financ. Anal. J.
Positive feedback investment strategies and destabilizing rational 103–106.
speculation. J. Finance 45 (2), 379–395. Kim, T., Ha, A., 2010 August. Investor sentiment and market anomalies. In:
De Roon, F., Eiling, E., Gerard, B., Hillion, P., 2012. Currency risk hedging: 23rd Australasian Finance and Banking Conference.
No free lunch. Available at SSRN: http://ssrn.com/abstract=1343644. Kirkpatrick II, C.D., Dahlquist, J.R., 2010. Technical Analysis: The Complete
Denis, D.J., Osobov, I., 2008. Why do firms pay dividends? International Resource for Financial Market Technicians, second ed. FT Press, New
evidence on the determinants of dividend policy. J. Financ. Econ. 89 Jersey.
(1), 62–82. Kurov, A., 2010. Investor sentiment and the stock market’s reaction to
Dennis, P., Mayhew, S., 2002. Risk-neutral skewness: Evidence from stock monetary policy. J. Bank. Finance 34 (1), 139–149.
options. J. Finan. Quant. Anal. 37 (03), 471–493. Lakonishok, J., Maberly, E., 1990. The weekend effect: Trading patterns of
Deshmukh, S., Goel, A.M., Howe, K.M., 2013. CEO overconfidence and individual and institutional investors. J. Finance 45 (1), 231–243.
dividend policy. J. Financ. Intermed. 22 (3), 440–463. Lakonishok, J., Smidt, S., 1988. Are seasonal anomalies real? A ninety-year
Driesprong, G., Jacobsen, B., Maat, B., 2008. Striking oil: Another puzzle? perspective. Rev. Financ. Stud. 1 (4), 403–425.
J. Financ. Econ. 89 (2), 307–327. Lamont, Owen A., Stein, Jeremy C., Aggregate short interest and
Dyl, E.A., Maberly, E.D., 1992. Odd-lot transactions around the turn of the market valuations (December 2003). Harvard Institute of Eco-
year and the January effect. J. Finan. Quant. Anal. 27 (4), 591–604. nomic Research Discussion Paper No. 2027. Available at SSRN:
Edwards, R.D., Magee, J., Bassetti, W.H.C., 2012. Technical Analysis of http://ssrn.com/abstract=569876.
Stock Trends. CRC Press, Boca Raton, FL. Lee, W.Y., Jiang, C.X., Indro, D.C., 2002. Stock market volatility, excess
Edwards, F.R., Zhang, X., 1998. Mutual funds and stock and bond market returns, and the role of investor sentiment. J. Bank. Finance 26 (12),
stability. J. Financ. Serv. Res. 13 (3), 257–282. 2277–2299.
Fama, E.F., Blume, M.E., 1966. Filter rules and stock-market trading. J. Bus. Lo, A.W., 2002. The statistics of Sharpe ratios. Financ. Anal. J. 58 (4), 36–52.
39 (1), 226–241. Lo, A.W., MacKinlay, A.C., 1990. Data-snooping biases in tests of financial
Fang, J., Jacobsen, B., Qin, Y., 2013. Predictability of the simple technical asset pricing models. Rev. Financ. Stud. 3 (3), 431–468.
trading rules: An out-of-sample test. Rev. Financ. Econ. 23 (1), 30–45. Lo, A.W., Mamaysky, H., Wang, J., 2000. Foundations of technical analysis:
Feldman, T., 2010. A more predictive index of market sentiment. J. Behav. Computational algorithms, statistical inference, and empirical imple-
Financ. 11 (4), 211–223. mentation. J. Finance 55 (4), 1705–1770.
Fisher, K.L., Statman, M., 2000. Investor sentiment and stock returns. Low, C., 2004. The fear and exuberance from implied volatility of S&P 100
Financ. Anal. J. 56 (2), 16–23. index options. J. Bus. 77 (3), 527–546.
Fleming, J., Ostdiek, B., Whaley, R.E., 1996. Trading costs and the relative Malek, L., 2000. The role of TED spread and confidence index in explaining
rates of price discovery in stock, futures, and option markets. J. the behaviour of stock prices. Amer. Bus. Rev. 18 (2).
Futures Mark. 16 (4), 353–387. Mayor, T.H., 1968. Short trading activities and the price of equities:
Fortune, P., 2000. Margin requirements, margin loans, and margin rates: Some simulation and regression results. J. Finan. Quant. Anal. 3 (3),
Practice and principles. New Engl. Econ. Rev. 19–44. 283–298.
Fosback, N.G., 1993. Stock Market Logic: A Sophisticated Approach to McQueen, G., Roley, V.V., 1993. Stock prices, news, and business
Profits on Wall Street. Dearborn Financial Publishing, Chicago, IL. conditions. Rev. Financ. Stud. 6 (3), 683–707.
56 J. Fang et al. / Journal of Behavioral and Experimental Finance 4 (2014) 25–56

Menkhoff, L., 2010. The use of technical analysis by fund managers: Solt, M.E., Statman, M., 1988. How useful is the sentiment index? Financ.
international evidence. J. Bank. Finance 34 (11), 2573–2586. Anal. J. 44 (5), 45–55.
Michaely, R., Thaler, R.H., Womack, K.L., 1995. Price reactions to dividend Stambaugh, R.F., Yu, J., Yuan, Y., 2012. The short of it: Investor sentiment
initiations and omissions: Overreaction or drift? J. Finance 50 (2), and anomalies. J. Financ. Econ. 104 (2), 288–302.
573–608. Summa, J., 2004. More mileage from margin. Tech. Anal. Stocks Commod.
Murphy, J., 1999. Technical Analysis of the Financial Markets: A 22 (3), 48–51.
Comprehensive Guide to Trading Methods and Applications. Penguin, Tabell, E.W., Tabell, A.W., 1964. The case for technical analysis. Financ.
New York.
Neal, R., Wheatley, S.M., 1998. Do measures of investor sentiment predict Anal. J. 20 (2), 67–76.
returns? J. Finan. Quant. Anal. 33 (4), 523–547. Vu, J.D., Caster, P., 1987. Why all the interest in short interest? Financ.
Nissim, D., Ziv, A., 2001. Dividend changes and future profitability. J. Anal. J. 43 (4), 76–79.
Finance 56 (6), 2111–2133. Wang, C., 2002. Information, trading demand, and futures price volatility.
Park, C.H., Irwin, S.H., 2007. What do we know about the profitability of Financ. Rev. 37 (2), 295–315.
technical analysis? J. Econ. Surv. 21 (4), 786–826. Wang, Y.H., Keswani, A., Taylor, S.J., 2006. The relationships between
Paulos, J.A., 2003. A Mathematician Plays the Stock Market. Basic Books, sentiment, returns and volatility. Int. J. Forecast. 22 (1), 109–123.
New York. Warther, V.A., 1995. Aggregate mutual fund flows and security returns. J.
Randall, M.R., Suk, D.Y., Tully, S.W., 2003. Mutual fund cash flows and Financ. Econ. 39 (2), 209–235.
stock market performance. J. Invest. 12 (1), 78–80. Whaley, R.E., 1993. Derivatives on market volatility: Hedging tools long
Reilly, F.K., Whitford, D.T., 1982. A test of the specialists’ short sale ratio. overdue. J. Derivatives 1 (1), 71–84.
J. Portfolio Manage. 8 (2), 12–18. Whaley, R.E., 2000. The investor fear gauge. J. Portfolio Manage. 26 (3),
Rhea, R., 1932. The Dow Theory. Barron’s, New York. 12–17.
Ring, J., 1974. Confidence Index, It Works Better at Tops than Bottoms. Woolridge, J.R., Dickinson, A., 1994. Short selling and common stock
Sanders, D.R., Irwin, S.H., Leuthold, R.M., 1997. Noise traders, mar- prices. Financ. Anal. J. 50 (1), 20–28.
ket sentiment, and futures price behavior. Available at SSRN: You, T., Holder, M.E., 2008. Can exchange seat prices predict financial
http://ssrn.com/abstract=39932. market volatility? J. Futures Mark. 28 (12), 1206–1221.
Scheinkman, J.A., Xiong, W., 2003. Overconfidence and speculative Yu, J., Yuan, Y., 2011. Investor sentiment and the mean–variance relation.
bubbles. J. Polit. Econ. 111 (6), 1183–1220. J. Financ. Econ. 100 (2), 367–381.
Schwert, G.W., 1977. Stock exchange seats as capital assets. J. Financ. Zakon, A.J., Pennypacker, J.C., 1968. An analysis of the advance–decline
Econ. 4 (1), 51–78. line as a stock market indicator. J. Finan. Quant. Anal. 3 (3),
Seneca, J.J., 1967. Short interest: Bearish or bullish? J. Finance 22 (1),
299–314.
67–70. Zweig, M.E., 1973. An investor expectations stock price predictive model
Shiller, R.J., 2000. Measuring bubble expectations and investor confi-
using closed-end fund premiums. J. Finance 28 (1), 67–78.
dence. J. Psychol. Financ. Mark. 1 (1), 49–60.
Simon, D.P., Wiggins, R.A., 2001. S&P futures returns and contrary
sentiment indicators. J. Futures Mark. 21 (5), 447–462. Further reading
Smith, R.D., 1968. Short interest and stock market prices. Financ. Anal. J.
151–154.
Solnik, B., 1993. The performance of international asset allocation White, H., 1980. A heteroscedasticity consistent covariance matrix
strategies using conditioning information. J. Empir. Finance 1 (1), estimator and a direct test of heteroskedasticity. Econometrica 48 (4),
33–35. 817–838.