HABIT FORMATION AND THE THEORY OF ADDICTION George Messinis The University of Melbourne Abstract. In the light of repeated rejections of the Hall (1978) version of the life cycle-permanent income hypothesis and other empirical puzzles, the habit formation hypothesis has increased in popularity since the 1980s. However, existing formulations of habit persistence do not always perform well empirically. This paper pursues two objectives: (i) to outline the habit persistence hypothesis, and (ii) to review the theory of addiction with a focus on issues of relevance to the theory of consumption. In the literature on addiction, two research traditions are discernible: rational addiction and myopic addiction. The former approach emphasises forward-looking behaviour and defines memory loss as a univariate process. The latter relies on multiple objectives and highlights the role of contractual behaviour. The paper argues that future research in consumption with habits ought to pay more attention to non-separabilities, allow for multivariate processes when modelling memory loss and consider rational habit modification. Keywords. Consumption; habits; addiction; myopia; habit modification 1. Introduction The life cycle-permanent income (LC=PIH) model of consumption predicts that risk averse consumers smooth consumption by saving in good times and by dissaving in recessions. When faced with a rising income profile, finite-lived consumers would also like to save when young and dissave when in retirement. The model was extended by Hall (1978) who showed that consumption ought to approximate a random walk process (assuming the real interest rate is constant and equal to the subjective rate of time preference). However, empirical research has identified three sets of puzzles: * * consumption follows current income too closely. In time-series data, changes in consumption correlate with changes in anticipated income (excess sensitivity)1 and are insufficiently sensitive to unanticipated changes in income (excess smoothness) Ð the Deaton paradox. In cross-section data, the life-cycle profile of consumption resembles that of current income (Deaton 1992);2 income growth Granger-causes saving with a positive sign, in high-growth East Asian countries and the US (Carroll and Weil 1994), and 0950-0804/99/04 0417±26 JOURNAL OF ECONOMIC SURVEYS Vol. 13, No. 4 # Blackwell Publishers Ltd. 1999, 108 Cowley Rd., Oxford OX4 1JF, UK and 350 Main St., Malden, MA 02148, USA. 418 * GEORGE MESSINIS saving behaviour by the elderly is highly heterogeneous: many reach retirement with little wealth while those with high income=wealth= education grow quite affluent (Browning and Lusardi 1996) but fail to adequately dissave prior to death (Kotlikoff 1989 Tachibanaki 1994 and Hayashi 1997).3 In response to these puzzles, research has explored various alternatives to the Hall (1978) model. These include liquidity constraints (Flavin 1985, Runkle 1991, Shea 1995 and Hayashi 1997), precautionary saving (Dynan 1993, Hubbard et al. 1994 and Pemberton 1995), omitted information (Pischke 1991, Goodfriend 1992 and Flavin 1993), bequests (Kotlikoff 1987 and Weil 1994) and habit persistence. Habit formation models of consumption have increased in popularity since the 1980s. These models have had some success in explaining excess smoothness in consumption (Ferson and Constantinides 1991), seasonality in consumption (Osborn 1988), saving behaviour in Japan (Christiano 1989), the positive Granger-causation from income growth to saving (Carroll and Weil 1994) and stock market behaviour (Campbell and Cochrane 1995 and Campbell 1996). However, existing models of habit persistence do not always perform well in empirical studies (Deaton 1992 and Muellbauer 1996). Moreover, they are at odds with recent evidence in Bauoumi and Koujianou (1990), Bauoumi (1993), de Brouwer (1996) and Olekalns (1997). Using post financial deregulation data, these studies fail to observe `excess sensitivity' and, thus, provide support for the liquidity constraints version of LC=PIH.4 This constitutes an implicit rejection of standard habit formation models since these models predict sluggish consumption even when capital markets are perfect. In order to enhance empirical performance of habit formation models, attempts have been made to integrate habit persistence and durability (Braun et al. 1993), disentangle habit formation, durability and time aggregation (Heaton 1995 and Ermini 1997), improve the specification of the habit stock (Abel 1990 and Campbell and Cochrane 1995), combine habit formation and income uncertainty (Alessie and Lusardi 1997) and improve the modelling of memory loss (Feichtinger 1993, Feichtinger, Hommes and Milik 1997 and Hindy, Huang and Zhu 1997). Habit persistence has also been prominent in microeconomic studies of addiction. Addiction is a strong form of habit formation and can be rational or myopic. Also, it can be commodity-specific (linked to the innate characteristics of a product) or generalised (linked to a lifestyle standard). The theory of addiction offers new insights into the specification of habit persistence and allows for diversity in consumer tastes. The objective of this paper is twofold: (i) to provide some intuition for habit formation models, and (ii) to review the theory of consumer addiction. The survey summarises major contributions from a wide range of perspectives. The paper is organised as follows. Section two outlines two alternative approaches to modelling rational habit formation in consumption. In section three, two alternative approaches to rational addiction are discussed. Section four focuses # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 419 on myopic habits arising from multiple centres of motivation. Section five makes suggestions towards a new model of consumption with habits, highlighting the need to consider rational habit modification. The final section offers some conclusions. 2. Rational habit formation Rational habits can be modelled either in terms of a stock of consumption capital that has its own law-of-motion (i.e., an additional constraint on the opportunity set of the maximising agent) or in terms of an endogenous discount function which leads to `accumulated' impatience. The former draws on optimal control techniques and the theory of economic growth, the latter builds on the idea of an endogenous rate of time-preference associated with the works of BoÈehm-Bawerk (1889), Fisher (1930) and Uzawa (1968). Rational habit formation involves time-dependent preferences where the marginal utility of consumption at time t depends on past levels of consumption. For brevity, consider the general specification max Et Ct ‡ i 1 X iˆ0 1 (1 ‡ ) i u(JCt ‡ i ) (1) subject to JCt ˆ C Ft ÿ bHt ˆ 1 X j Ct ÿ j ÿ b jˆ0 1 X as C Ftÿ s j ; b; as > 0 (2) sˆ1 and the lifetime budget constraint 1 X iˆ0 1 (1 ‡ r) i Et Ct ‡ i ˆ At ‡ 1 X iˆ0 1 (1 ‡ r) i Et Yt ‡ i  Wt (3) Et is an expectations operator,  is the pure rate of time preference, u (
) is a felicity function, P Ct is consumption expenditure at time t, Ht is the habit stock, CFt ˆ j 1 ˆ 0  jCt ÿ j is the flow of services, JCt stands for relative consumption, r is the real interest rate (a constant), A is net worth and Wt stands for wealth. Suppose that utility is quadratic, expectations are formed rationally and r ˆ . Then, the model nests the following hypotheses: 1. 2. 3. 4. the random walk hypothesis of Hall (1978), when b ˆ 0 and CFt ˆ Ct; the pure durable hypothesis of Mankiw (1982), when b ˆ O and CFt 6ˆ Ct; the standard habit formation model, when b > 0 and CFt ˆ Ct, and Ermini's (1997) model of habits with durability, when b > 0 and CFt 6ˆ Ct. We focus on the third model to gain some intuition into habit formation. Past consumption of C raises the consumption standard, Ht, and makes a unit of current consumption less enjoyable. Contrary to the public image of the new rich # Blackwell Publishers Ltd. 1999 420 GEORGE MESSINIS indulging in instant gratification, the model suggests that the marginal rate of substitution between present and future consumption, MRSPF ( ˆ UP=UF), will be higher for the long-term rich. Thus, those who have been accustomed to a relatively low living standard in the past will be more willing to postpone consumption. It is instructive to show how habit persistence can explain several of the empirical puzzles summarised above. For example, both excess sensitivity and excess smoothness can be attributed to habit formation.5 First, suppose that CFt ˆ Ct and that habit formation follows a geometric process, i.e., as ˆ as. Then, the problem (1) ± (3) has the closed-form solution
Ct ‡ 1 ˆ (1 ‡ b)a
Ct ‡ ut ‡ 1 ÿ aut (4) where ut is an error term and consumption is an ARIMA (1,1,1) process. Equation (4) explains why Hall's (1978) random walk hypothesis will be rejected against the excess sensitivity alternative when habit formation is ignored. Note that
Ct will depend on both ut and
Ct ÿ 1, the latter will depend on ut ÿ 1 and
Ct ÿ 2, and so on. Thus, if
Ct ÿ 1 is omitted,
Ct will correlate with expected income at time t, to the extend its past innovations systematically relate to innovations in permanent income. If habit formation is important, the econometrician will also find consumption to be excessively smooth. To see this, we simplify the model further. Suppose that at time t the standard of consumption is only last year's consumption, Ct ÿ 1. That is, relative consumption is of the form JCt ˆ Ct ÿ bCt ÿ 1 with b > 0. Given linearity in marginal utility, rational expectations and r ˆ , the solution to (1) is EtJCt ‡ 1 ˆ JCt (5) Muellbauer (1988) has shown that JCt ˆ r 1‡r HWt (6) where Wt is the resources that can sustain lifetime relative consumption, JCt. If we substitute Ct for JCt ‡ bCt ÿ 1 in (3), we get Wt ˆ 1 X iˆ0 1 (1 ‡ r) i Et [JCt ‡ i ‡ bCt ‡ i ÿ 1 ] (7) By expansion and substitution, we reach Wt ˆ HWt ‡ bCt ÿ 1 (8) where ˆ 1 X iˆ0 0 @ # Blackwell Publishers Ltd. 1999 b 1‡r 1i A ˆ 1‡r 1‡rÿb ; HWt ˆ 1 X iˆ0 1 (1 ‡ r) i Et JCt ‡ i HABIT FORMATION AND THE THEORY OF ADDICTION 421 Next, we use this result to substitute out JCt and HWt in (6) to get an expression for optimal consumption: Ct ˆ r (1 ‡ r) Wt ‡ b 1‡r (9) Ct ÿ 1 Now it is easy to express (9) in terms of permanent income, Y pt : 0 1 1 p b b b AY pt ‡ Ct ÿ 1 ˆ @ 1 ÿ Ct ÿ 1 Ct ˆ Y t ‡ 1‡r 1‡r 1‡r (10) where the second equality shows that consumption is a weighted average of permanent income and past consumption. Taking expectations, it is easy to show that Ct ÿ Et ÿ 1 Ct ˆ 1 (Et ÿ Et ÿ 1 )Y pt ˆ 1 t (11) This result demonstrates that, if habits are important, the LC=PIH model overstates the response of consumption to innovations in permanent income, since > 1. Hence, irrespective of the order of integration in labour income, when the habits are ignored in testing the two hypotheses, consumption will exhibit `excess smoothness'. The model can also explain the stylised fact of a positive Granger-causation between growth and saving. In a growing economy, it is sensible to expect income to grow. In this scenario, the model predicts growth in consumption. This is due to three factors. First, the consumer is conscious of the fact that higher consumption now will lead to a higher future marginal utility in the future. Second, assuming the initial endowment for H is low (relative to permanent income), consumption has to grow over time in order to offset a rising Ht if JCt is to remain constant over time. Finally, in a growing economy the number of new rich is growing while that of the new poor is shrinking; i.e., Ct grows faster than Ht (Becker 1992). The net effect is that MRSPF declines with Ht and the typical consumer appears to be prudent. Although the above model greatly enhances intuition of rational habit formation, it is by no means the only approach to modelling rational habits. An alternative approach originates in BoÈehm-Bawerk (1889) and Uzawa (1968) who perceived the rate of time-preference to be endogenous. Following BoÈehmBawerk (1889), a century-long debate has evolved regarding the concept of pure rate of time preference, .6 Conventional wisdom has settled with the view that  is positive.7 Research also has explored the possibility that  is a non-constant. This field of research builds on recursive utility functions that maintain consistency in intertemporal planning. Here, time-nonseparable but consistent preferences arise from endogenous time preferences. In discounting the future, the consumer takes into account the cumulative effect of past utility or consumption. # Blackwell Publishers Ltd. 1999 422 GEORGE MESSINIS A number of models have been proposed. The model by Shi and Epstein (1993) (SE) is particularly appealing for it provides flexibility in the representation of habit formation. They suggest a recursive utility function of the following kind: 1 U(C) ˆ u(C)e 0 ÿ R1 0 (Z)dsdt (12) where Z_ (t) ˆ k[C (t) ÿ Z (t)], Z (0) ˆ Z0 å 0, k>0 (13) and t C(s)e k(sÿt)ds Z(t) ˆ k (14) ÿ1 where C (t) is a consumption path, Z (t) is the habit variable,  denotes the rate at which Z (t) adjusts to past consumption and u (
) and  (
) are the instantaneous utility and discounting functions respectively (both are real valued and twice differentiable). The properties of  (Z) is a central issue. Although the consensus is for a concave  (
), the sign of  0 (
) remains controversial. The traditional view is one of an inverse relationship between wealth and impatience. This assumption seems reasonable in as far as low income groups are concerned. However, habit persistence suggests a positive association between consumption and consumer impatience (Obstfeld 1990). Consumers with a standard of living well beyond subsistence become accustomed to a consumption lifestyle which, in turn, determines their rate of time preference. This effect is mainly due to Fisher (1930). In his comments on the phenomenon of `social mobility', he referred to habit formation as a process which leads wealthy people to a life of extravagance and assists those accustomed to a `simple and inexpensive' lifestyle towards a substantial accumulation of wealth; rich and poor can both become frugal or improvident since their rate of time preference changes over the life cycle.8 Returning to the SE model, the generality of the model rests on the k factor. When k ˆ  0 ˆ 0 there are no habits, the rate of time preference is constant and U reduces to a time-additive function. The Uzawa (1968) model is also encompassed; that is, when  ! 1 ,  (Z) ˆ  (C) and the rate of time preference only depends on current and future consumption. Time-nonseparability in the SE model is summarised in the (local) rate of time preference function,  (Z, C), along a constant consumption path. The authors evaluate both Z and C to be positive suggesting that patience increases with past and current consumption. Further, the forward-looking nature of the model rests # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 423 on the complementarity between current and future consumption:  (Z, C) positively relates to consumption at time t in the future 0 1 d (Z; C) @ i:e:; > 0 A: dC(t) The model predicts that all individuals will hold positive assets in the long-run (Epstein 1987 and Obstfeld 1990). Here, relative impatience simply implies a lower steady-state consumption path.9 This prediction is intuitively appealing and consistent with stylised facts about the rich. On the other hand, the assumption of  (
) 0 > 0 seems counter-intuitive for the lower income groups. 3. Rational addiction 3.1. The Becker and Murphy model Becker and Murphy (1988) draw on the literature of rational habit formation (Stigler and Becker 1977 and Iannacone 1986) to model consumer addiction as a rational process. The addict is rational and foresighted. In maximising utility, the consumer takes into account the future consequences of past and current consumption of addictive substances. More formally, addiction is explained in terms of a stock variable, S, entering utility as an argument. Past consumption of the addictive good, C, leads to the accumulation of the stock (i.e., commodityspecific addiction). In essence, an addiction is a strong habit effect which can be destabilising with respect to current consumption. Formally, the consumer lives an infinite life and maximises 1 e ÿt u[C(t); S(t); N(t)]dt (15) S_ (t)ˆC (t) ÿ S (t) (16) 0 subject to and 1 1 e ÿrt [N(t) ‡ pC (t)C(t)] Æ A0 ‡ 0 e ÿrt w[S(t)]dt (17) 0 where u (
) is momentary utility (concave and time-separable in C, N and S), N (t) is consumption of nonaddictive goods,  is the rate of time preference, r is a constant real interest rate,  is the depreciation rate, pC (t) is the money price of C (t), A0 is the initial value of assets and w (
) is earnings, being a concave function of the state variable, S (t). Potential addiction is associated with adjacent complementarity. The latter is present when a positive correlation exists between current and future consumption # Blackwell Publishers Ltd. 1999 424 GEORGE MESSINIS of C or when the marginal utility of consumption from a change in S, aCS > 0, rises above the shadow price of C.10 By assuming a quadratic utility function and substituting for the optimal value of N, Becker and Murphy (hereafter, BM) arrive at the optimal condition C (t) ˆ S (t) ‡ 1[S (t) ÿ S ] (18) where S is the steady state value of S and 1 is the smaller characteristic root of a second order Euler equation. It follows that potential addiction is present when ( ‡ 2)UCS > ÿUSS and the addiction is destabilising if US > , given that C ˆ S in the steady state. An indispensable feature of this model is the existence of two steady states, S0 and S1 , where the lower state, S0 , is unstable due to a high degree of adjacent complementarity.11 Also germane to the model is the relative position of the initial stock, S0. When S0 is greater (smaller) than S0 the optimal consumption path is one of addiction (abstention).12 In addition to S0, variations in the price of C can also induce individuals to reverse their consumption decisions. For example, if the current price of cigarettes rises, a smoker who would otherwise be moving towards addiction may quit smoking altogether.13 This is because the rise can cause the unstable steady state to increase above S0.14 Finally, there is an attempt to explain how consumption of C first arises. Here the model relies on exogenous shocks. These are large changes in the mental state of a potential addict. Triggered by highly stressful events, these changes are thought to raise the demand for addictive substances and, thus, explain how individuals with no experience in C become addicted. The effect is consistent with observations of a positive correlation between addiction and stress. In order to account for such shocks, Becker and Murphy revise their differential equation describing S to include an endowment stock, Z (t), which adds to the marginal utility of consumption of C as does S. Criticism of the BM model has centred on three issues. The first refers to the idea of addiction as intended behaviour. In the model, the consumer is an optimiser who has no regrets about previous decisions. However, this idea seems to contradict the stylised fact that many drug addicts, smokers and alcoholics repeatedly regret their dependence on these substances and, in that sense, they cannot be `happy' addicts (see Winston 1980 and Akerlof 1991). Moreover, many addicts seem unable to break their addiction when they try.15 A second criticism relates to the failure of the model to explain casual consumption of addictive goods for extended periods. A final criticism relates to the stability of the system and the interpretation of empirical estimates of long-run demand elasticities. The BM model assumes an infinite planning horizon. Ferguson (1996) has shown that in a simple BM model, modified for a finite planning horizon, the economy will most likely diverge away from the long-run equilibrium if the transversality condition is to be satisfied.16 Also, in the absence of a market for S, there exists no mechanism for the system to jump back to the stable arm. As a result, aggregate consumption of addictive goods will tend to be explosive and long-run elasticity estimates will be unreliable. # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 425 3.2. Regret and learning Orphanides and Zervos (1995) (OZ) address some of the criticisms directed at the BM model. In OZ, individuals are uncertain about their potential to become addicted. They hold subjective beliefs which they update through a Bayesian process of learning. The update is based on information derived from past consumption of addictive products. Addiction originates from mistaken beliefs regarding one's likelihood of becoming an addict. Potential addicts risk becoming addicted because there is certainty about the benefits of immediate consumption but only a probability of harmful effects being realised in the future. In this model, the finite-lived consumer maximises Et 1 X (1 ‡ ) ÿi [u(Ct ; Nt ) ‡ t (Ct ; St )] (19) Nt Å Yt ÿ pCt; (20) iˆ0 subject to C t, N t å 0 St ‡ 1 ˆ Ct ‡ (1 ÿ )St ˆ Ct ‡  St (21) and t ˆ ( 1; with probability of (St ) 0; with probability of 1 ÿ (St ) (22) where  and  are as in the BM model, Ct is the addictive good, Nt is the nonaddictive good,  (
) is the disutility effect of past consumption of Ct, u (
) is a the felicity function, St is the habit stock, Yt is income, p is the price of a when the price of Nt is the numeraire,  takes the value of one when the consumer is an addict and of zero when nonaddict and  (
) is the probability of harm caused by addiction. Instantaneous learning is disallowed by introducing a random variable, t 2 (0, 1), describing the harm caused by Ct. The following assumptions apply: 1. the probability of harm occurring depends on past consumption, St; 2. both u (
) and  (
) are twice differentiable; the former is strictly concave in Nt and Ct, the latter is concave only in Nt; 3. Ct is a complement to St and Nt (i.e., both partial derivatives uAS å 0 and AC å 0), and 4. the probability of harm is increasing in S, S å 0. Whether an individual with no previous experience in Ct will try the addictive good it depends on a subjective prior distribution, P0 ˆ P ( ˆ 0), and a critical P 2 (0, 1). The decision should be in the affirmative if the person is believed to be # Blackwell Publishers Ltd. 1999 426 GEORGE MESSINIS a nonaddict (P0 å P ) and in the negative if he is believed to be an addict (P0 < P ).17 The Bayesian rule of updating the initial prior is assumed to be 8 > < Pt ; when  ˆ 0 t (23) P t ‡ 1 ˆ 1 ÿ t > : 0; when t > 0 where t ˆ  (St) (1 ÿ Pt). By using the budget constraint to substitute for N, w (C) ˆ u (Y ÿ pC, C), a unique value function is derived: 2 2 3 1 V(  S ‡ C; 0) 5 V(S; P) ˆ max 4 w(C) ‡ 4 (C; S) ‡ C 1‡ 2 33 (1 ÿ ) P 55 V4   S ‡ C; (24) ‡ 1‡ 1ÿ In (24), the last two terms within the outer brackets reflect the expected effect of consumption on future utility. Assuming only harmful addiction, the consumer expects with probability  that utility will be adversely affected by  (C, S) and revises the prior to zero and the discounted value to 1= (1 ‡ )V ( S ‡ C, 0). Further, the consumer anticipates no harmful effects with a probability (1 ÿ ), updates  upwards and revises the discounted value to 1= (1 ‡ )V[ S ‡ C, P= (1 ÿ )]. Optimal consumption follows a different path depending on whether the consumer is certain about . When there is certainty that the agent is not a potential addict (i.e., P ( ˆ 0) ˆ 1), the steady state is S ˆ C = (1 ÿ  ) where C > 0 is the static optimum level of C that satisfies the first-order condition, w (C ) ˆ 0. In this case, the agent can become addicted if S is greater than a critical value, Sc. With uncertainty about , on the other hand, only those who experiment with C are at risk of becoming addicted. Addiction is possible when, in failing to observe the harmful effects of past consumption, individuals assume that t ˆ 0, update Pt ‡ 1 upwards and are, thus, misled into underestimating the risk of addiction.18 Contrasted with the BM model, addiction in this case is the unintended outcome of uncertainty and erroneous underestimation of the harm caused by the consumption of addictive substances. Therefore, addicts are surprised to find themselves drawn into an addiction and they regret having been overly optimistic in their assessment of the initial prior. The model is a big step towards realism since it is consistent with experimental evidence of violations to the Bayesian Theorem whereby individuals appear to give insufficient weight to prior information (Earl 1990). An example would be young people ignoring parental advice regarding drugs or financial matters. Also, the model goes further: (i) it acknowledges the role of past consumption and public information in the # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 427 formation of subjective beliefs; (ii) it assumes that risk taking and misjudgment suffice to make addicts out of inexperienced consumers (peer pressure is part of the misinformation problem); (iii) it allows for prolonged but moderate consumption of goods such as coffee and alcohol, and (iv) it explains why many abstainers who have never experimented with addictive goods overestimate the harmful side effects of goods such as alcohol. Although the OZ model represents a considerable improvement, it raises a number of questions. For example, how are we to distinguish between those consumers who are certain about  and those who are not, or between those who observe the harmful effects of addiction and those who fail to do so. Further, is casual consumption of addictive goods due to one's ex ante self-knowledge of not being a potential addict or is it due to learning through experience? On this last question, it seems that the latter explanation is more in tune with the central idea in the OZ model: rather than prior knowledge, it is experience in addictive consumption that teaches people to manage their addictions. Finally, there is the additional challenge of developing an empirical framework for the OZ model that will make it distinguishable from the BM model. 4. Myopic habits and multiple objectives The second approach to explaining habitual behaviour centres on the idea that consumers suffer from myopia and, thus, do not consider the effects of addiction on future utility. In the past, the modelling of myopic habits was based on ad hoc assumptions without sound theoretical foundations. As a result, economists have tended to dismiss myopic habits. Muellbauer (1988) is an exception. Using the same notation as in section 2, myopic habits can be specified as follows. The consumer maximises max Et Ct ‡ i 1 X iˆ0 1 (1 ‡ ) i u(JCt ‡ i ); u 0 > 0; u 00 < 0 (25) subject to JCt ‡ i ˆ C Ft‡ i ÿ bC Ftÿ 1 ˆ 1 X j Ct ‡ i ÿ j ÿ bC Ftÿ 1 b>0 (26) iˆ0 In contrast to (2), the consumption standard for period t ‡ i now depends on consumption in period t ÿ 1 and not in t ‡ i ÿ 1. Muellbauer assumes CFt ˆ Ct and uses quarterly US data for non-durables to test the model against the rational habits hypothesis. Overall, his findings indicate that myopic habits are important. In recent years, this field of research has recorded some progress towards better specifications for myopic habits. Much attention has been given to the proposition that myopia originates in multiple centres of motivation, an idea first introduced in economics by Georgescu-Roegen (1954). Here, consumer behaviour is no more driven by the single ultimate objective of lifetime utility maximisation. More # Blackwell Publishers Ltd. 1999 428 GEORGE MESSINIS precisely, the decision making unit divides into a number of control centres. These centres maintain some autonomy which, under certain conditions, can give rise to conflict and myopic decisions. So far, the focus of research has been on theoretical constructions rather than on empirical work. 4.1. Non-constant stable preferences In response to Stigler and Becker (1977), Winston (1980) maintains that conflict is an essential feature of addiction. In order to integrate conflict, Winston introduces a dual motivation structure. The principal assumptions of the model are as follows: 1. the individual derives utility from the consumption of two composite commodities: M (addictive) and Z (nonaddictive); 2. addiction is reflected in the household production functions: Mj ˆ M (tmj, Xmj), Zj ˆ Z (tzj, Szj) and Szj ˆ S (Mj, Mj ÿ 1, ...) where time tmj and goods Xmj are inputs used for the production of a unit of commodity M within a period j, and Szj is the stock of human capital necessary for the production of good Z. Intuitively, consumption experience in M, say listening to music, leads to the accumulation of a skill, say the skill of music appreciation (reflected in S). In turn, this skill contributes to the production of another nonaddictive commodity, Zj, which is also a choice variable in the utility maximisation problem; 3. within a period j, the consumer oscillates between two stable states of preferences, described by two separate utility functions, U ˆ U (M, Z) and U ˆ U (M, Z). The only difference between U and U is that the marginal utility of consumption of M, UM, is greater in state U than in state U. That is, at a certain time of the day the consumer is more susceptible to addiction;19 4. for simplicity, the consumer is always certain about the structure of U (
) and U (
) and about the frequency in which each state dorninates the self, and 5. the consumer makes long-term plans according to a long-run utility function, IU ˆ IU (M, Z) which is the weighted sum of the two short-run preferences. This ensures that the non-constancy of short-run myopic preferences does not impact on optimal choice in the long-run. Given assumption (3), Winston shows that, at the point of zero consumption for M, it is always true that the marginal rate of substitution between M and Z in preference state U, MRSMZ ˆ UMj=UZj, is smaller than that in state IU and even smaller than that in U , i.e., MRSMZ < MRSMZ < MRSMZ . This condition gives rise to multiple and contradictory decision rules in intertemporal consumption of good M. To clarify, suppose that the relative shadow price ratio at time j is MZ ˆ PMj=PZj. Note first that when MRSMZ > MZ (MRSMZ < MZ) the optimal decision is unequivocal consumption (abstinence) of M Ð the consumer will not hesitate to increase (avoid) the consumption of M.20 If, however, MRSMZ < MZ < MRSMZ , short-run prefer# Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 429 ences U (
) will call for consumption when, in fact, it is optimal in the long-run to abstain from M. Also, if MRSMZ < MZ < MRSMZ, preferences governed by U (
) will make the consumer reluctant to consume even if it is rational to do so according to long-run preferences. In these last two cases, the consumer faces contradictory optima and will be suspended between a myopic state of preferences and a long-run code of behaviour.21 This conflict, however, can always be contained through actions that increase MZ or decrease MRSMZ , assuming MZ is known. Overall, Winston (1980) deals simultaneously with two major issues. First, it integrates conflict in the optimisation process. Here, conflict originates in multiple preference states and not in uncertainty or information problems. Second, it allows for a variety of strategies in managing addiction. One such strategy may involve within-period separabilities. Recall that the disutility associated with addiction operates through its effect on non-addictive goods (assumption 2). That is, consumption of addictive substances imposes a harmful externality on other goods. For example, smoking reduce one's capacity to enjoy other goods. This raises the question as to whether beneficial activities such as jogging can help one man age addiction to smoking or alcohol. Winston, thus, alludes to a model where externalities moderate dependence on addictive substances by decreasing MRSMZ . An alternative strategy is to increase the shadow price of addiction which causes MZ to rise. These include precommitment to a savings plan such as Christmas Clubs and `anti-market' products (e.g. dieting schemes, anti-smoking devices and consumer credit counselling). However, serious conceptual problems remain in Winston (1980). One is that the model fails to elucidate on how the conflict is resolved; it is not quite clear which state of preference gains command over the self when decisions are made. Although at a particular point in time the individual is aware of both short-run and long-run states, it is implicit that only myopic preferences dominate in the decision making process. Further, Winston never explores how the decision to precommit is made. While motivation requires a long-run perspective by the consumer, the actual decision to precommit implies that the consumer is able to abide by the long-run preference state. This, however, seems to contradict the assumption of myopic preferences having full command over the self. Finally, the complexity of the Winston's model implies that it has little to offer empirical research. This is mainly due to a lack of data regarding (i) the frequency in which the individual lives under each myopic state and (ii) the shadow price of addictive goods. 4.2. A tripartite brain structure Taylor (1992) also rejects the view that the individual has a single objective. He perceives consumption as the outcome of decisions made in three distinct centres of motivation; that is, there is not a single utility function which the consumer optimises. Taylor draws on developments in neuro-science that point to a threebrain structure of the self. The three brains, alpha, beta and gamma, pursue # Blackwell Publishers Ltd. 1999 430 GEORGE MESSINIS separate but complementary goals. The alpha brain seeks to fulfil basic physiological needs, the beta brain forms tastes, habits, acts efficiently and learns only from experience. The gamma brain is intelligent, monitors the activities of the other brains, forms expectations and can influence the other brains. A generalised model is proposed in which temporal utility is a function of the disparity between actual and desired consumption, the latter being a function of a desired stock variable, S . This state variable is made up of two components: a physiological reserve and a psychological stock associated with a minimum level of comfort. Actual consumption, on the other hand, is the net outcome of an `opponent process'. An initial process describes the comfort (utility) derived from various levels of consumption of a product. With a lag, consumption gives rise to a second effect which yields discomfort (disutility).22 When the level of discomfort is strong enough to exceed a certain threshold, the beta brain triggers a repeat (redose) of the same consumption activity. Redosing leads to addiction, for the state variable describing the discomfort effect relates positively to the number of repeats. A sophisticated gamma brain together with a hierarchy of needs23 ensure that consumer choice is exercised on products associated with `creativity' and individual expression. This rests on the assumption that these products Ð mainly luxuries Ð produce a level of discomfort that is smaller than the threshold level mentioned above and are, thus, less likely to be addictive.24 Therefore, habits and consumer choice coexist. Although neither habits nor opponent processes can be eliminated, a purposeful gamma brain will take action to counter an addiction if it causes the overall state of well-being to fall below a critical value. In such a situation the agent may be observed to experience conflict, reflecting the gamma brain's dissatisfaction with the habitual existence of alpha and beta brains. The model focuses on decision making processes rather than on outcomes and provides a plausible rationale for habit persistence and error-correction models. Moreover, it goes beyond myopic habits to incorporate consumer choice and forward looking behaviour without necessarily relying on rational addiction as in the Becker and Murphy model. However, Taylor's model is incomplete with respect to the specification of the gamma brain and the interaction of the three brains. Taylor's primary concern is a dynamic theory of demand that is compatible with new insights in the theory of decision making. In particular, he speculates that opponent processes and multiple centres of motivation imply lexicographic and non-transitive preferences. However, there is little information on how these preferences relate to the model. Also unspecified is the reaction of the gamma brain to harmful addiction. Finally, there is the difficulty of measuring unobservable variables such as `hierarchical' needs and `comfort threshold'. 4.3. Self-control and mental accounting The conflict between short-run and long-run preferences can also arise from selfcontrol problems. Influenced by Plato's Republic, J. S. Mill, was the first # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 431 economist to perceive the self as an acrobat seeking to strike a balance between `mind' and `body' or `will' and `desire'. In departing from Plato, however, Mill (1859, 1863) considered desires and impulses to be vital qualities of a perfect human being as long as they are counterbalanced by a strong will. However, this balance cannot be taken for granted and any imbalance can only be attributed to an `infirmity of character'. Mill's contribution to the self-control literature consists of three postulates: (i) self-control problems are quite distinct from information problems; (ii) time and society teach individuals to exercise willpower; and (iii) society as a whole and individuals over their life-cycle can oscillate from states of excessive impulsiveness to states of `deficiency' in personal preferences and impulses, the latter being due to a rigid obedience to rules and customs. In contemporary economic theory, it was BoÈhm-Bawerk (1889) who first made an explicit reference to lack of willpower as one of the three main causes of a positive time preference.25 Self-control also featured in Fisher's (1930) work on intertemporal consumption who added habit, life expectancy, altruism and `fashion' to the list. Research interest in self-control has revived recently (Elster 1979, Schelling 1984b, Hoch and Loewenstein 1991, Frank and Hutchens 1993, and Archibald 1994a, 1994b). Shefrin and Thaler (1992) (ST) attempt to introduce self-control in consumer behaviour. They perceive the individual as a two-self entity. One part is the `planner', prudently concerned with lifetime utility; the other is the `doer' who is `pathologically' myopic but has full control of the self in the current period. Since the `planner' does not have control over the `doer', conflict arises. The `doer' is oriented towards maximum current consumption; the `planner' seeks to maximise lifetime consumption. In order to restrain the doer, the `planner' exerts willpower effort to reduce present consumption which leads to a psychic cost or utility loss. Formally, the total utility for the individual, at time t, will be Zt ˆ U (Ct) ‡ W [ (Ct , Xt)] (27) where U (
) is a normal utility function and W [
] describes the `psychic cost' of willpower effort, , required to induce the `doer' to limit consumption to the optimal level Ct when a set of feasible consumption choices, Xt, is available at period t. If we replace Xt by current income, Yt, ignore the time subscripts and denote partial derivatives by a subscript, equation (27) has the following properties: # Blackwell Publishers Ltd. 1999 C < 0 (28) W < 0 (29) Zcc < 0 (30) D ˆ Zc ÿ Zc > 0 (31) ZY < 0 (32) 432 GEORGE MESSINIS Inequality (28) states that less willpower effort is required to adjust to a higher optimal consumption path. According to (29), this adjustment produces extra utility. Further, (30) ensures that the utility function is concave in C and (31) depicts the net marginal cost of using willpower which is positive when temptation is present (i.e., when Yt ÿ Ct > 0). Since temptation is directly proportional to current income, Yt, a utility loss is involved when current income increases, (32). For illustration purposes, consider a two-period horizon where the consumer maximises V ˆ Z (Ct, Yt) ‡ EtU (Ct ‡ 1) ˆ U (Ct) ‡ W (Ct, Yt) ‡ EtU (Ct ‡ 1) (33) Subject to a standard budget constraint, an optimal rule can be derived for the ST model UCt UCt ‡ 1 ˆ ZCt Et UCt ‡ 1 ˆ UCt ‡ WCt Et UCt ‡ 1 ˆ 1‡r 1‡ (34) where Et, r and  are as defined in previous sections. Note here that the rational expectations version of LC=PIH is only a special case when self-control is assumed constant (i.e., WCt ˆ 0). Given (32), (34) shows that an increase in income at time t would increase the marginal utility of consumption and would, thus, motivate the consumer to increase current consumption. What complicates matters further, however, is the ability of the `planner' to take precautionary steps to restrain the `doer' self. In contrast to most of the previous models, the rational consumer is not a passive player in the ST model. The `planner' may be able to participate in a contractual savings plan to reduce willpower disutility and realise a long-term consumption plan. In order to justify this, the self-control literature often refers to Homer's Odyssey and Ulysses's journey back home to Ithaca. In Troy he had made a reputation for his cunningness, but in order to avoid being captivated by the singing of the Sirens he blocked his men's ears with wax and bound himself to the mast. Precommitment is thus necessary but the question of what prevents the `doer' from squandering their savings when the plan matures requires an answer. Shefrin and Thaler's response is the concept of `mental accounting': people code their resources into three different mental accounts: current income (Y), asset wealth (A) and future wealth (F). These are internal constraints which make some wealth accounts less fungible than others. More precisely, access to A and F by the `doer' is relatively harder. This is expressed in terms of differential marginal propensities to consume: MPCY > MPCA > MPCF. Being aware of these constraints, the rational `planner' self will try to change their wealth composition by committing to a savings plan such as Christmas Saving Clubs and income retirement plans. Mental accounting, however, does not only imply pro-active behaviour such as precommitment. It also predicts that institutional or life-cycle factors can substantially modify consumer behaviour. Shefrin and Thaler (1998, 1992) go to some length to highlight the importance of Individual Retirement Accounts (IRAs), bonus income, and housing wealth. For # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 433 example, bonus income is perceived to enter the A account and not the current income account. It follows, saving rates in countries where bonus income is an important component of household income will be expected to be relatively high.26 Also, home ownership seems to be a highly important life-cycle event. Housing wealth is seen to be less tempting than the assets account.27 Shefrin and Thaler never provide a rationale for this but Thaler (1992, 1994) has highlighted mortgage credit as an important feature of home ownership. The ST model predicts that consumption will be hypersensitive to changes in current income (expected or not). This is also attributed to mental accounting. Recall that each wealth account has its own marginal propensity to consume attached to it. Suppose that current income falls due to a recession. Although consumers may be interested in consumption smoothing, they will be reluctant to withdraw on their assets or future income since the marginal propensity to consume is lower for these accounts than for current income. This also explains why they would refrain from borrowing. Thus, consumption will follow current income. In boom periods, on the other hand, temptation increases and, given equation (32), current consumption will tend to rise if condition (34) is to be satisfied. Precommitment and mental accounting, however, do not constitute a watertight argument. If ultimate control of the self rests with the `doer', as suggested by Shefrin and Thaler, what induces the `doer' to approve of the portfolio shift or what prevents the `doer' from changing the composition of wealth? If it is a principal-agent problem, what is the incentive payment? Also, the idea that mortgage credit constitutes a saving device creates an identification problem: if in post-deregulation consumer behaviour has changed, is it due to a relaxation of liquidity constraints or to a higher demand for contractual savings? Shefrin and Thaler insist that mortgage credit is always demand-driven but this is of little comfort to applied economists. Yet, all of the above problems may be resolved if we assume that the `doer' self is an addict who acts according to a habit effect produced by past consumption of particular goods. In fact, Shefrin and Thaler repeatedly refer to `bad' and `good' habits in order to explain why the retired do not relax their rules to adequately dissave in retirement28 or to explain why self-control problems are overcome. In this context, there would be diversity in consumers' response to an increase in current income. Moreover, mortgage credit or subscriptions to Christmas Clubs would not just be devices toward target savings but also habit-modifying. The reinterpretation of the self-control problem as a habit effect can also be deduced from the Odyssey. We ask the question: what inference can be drawn from the fact that Ulysses listened to the Sirens instead of blocking his ears with wax? One possible interpretation is that through this experience Ulysses learnt to face temptation in the future. This interpretation is consistent with Odyssey: Ulysses survives the final adventure but all his men die as they fail to resist temptation.29 Therefore, a more coherent rationale can be provided for a habit formation model in which some consumer goods are habit-forming while others cause the habit stock to deplete. For example, one can envisage non-durables as habit# Blackwell Publishers Ltd. 1999 434 GEORGE MESSINIS forming and services from housing=durables or mortgage credit as habitdepleting. This would provide a new insight into the stylised fact of home owners not withdrawing on their housing equity (Venti and Wise 1989).30 This new interpretation of the role of housing removes the link between precommitment and mental accounting and facilitates a more parsimonious model that can include liquidity constraints. 5. Directions for future research In this section, we draw on the theory of addiction to suggest a number of strategies towards a better specification of habits in consumption. We highlight ideas which are most useful towards a new approach to habit formation: rational habit modification. The first idea deals with consumer rationality. The proposition that the typical consumer has multiple personalities or multiple centres of motivation is intuitively appealing. However, this literature has yet to develop a coherent theory of decision making. For example, in all three of the models considered, it is unclear which part of the self decides what. Also unresolved is the question of how the rational self in Winston (1980), the gamma brain in Taylor (1992) or the `planner' self in Shefrin and Thaler (1988) induces the myopic self into a contract.31 Thus, at this stage, the assumption of consumer rationality seems quite reasonable. The second idea originates in Winston (1980). Recall that, in his model, addiction to a commodity M impacts on the utility derived from commodity Z. Although this idea has attracted little attention in the literature, it is an important proposition since it alludes to a model where (i) utility is nonseparable, and (ii) nonseparabilities play a role in the withdrawal from addiction.32 This may be particularly important in the light of time-series evidence rejecting the assumption of utility being additively separable in durables, services and non-durables (Startz 1989, Fauvel and Samson 1991 and Patterson 1994). The third idea relates to the specification of memory loss. As noted in the introduction, recent research on rational habits has also focused on this issue. Most notable is the work by Feichtinger, Hommes and Milik (1997) and Hindy, Huang and Zhu (1997). They extend the BM model by treating (i) addiction or reinforcement, and (ii) satiation or withdrawal as two distinct processes. They adopt a two-state approach in which the consumption of a single good gives rise to two stocks; one produces addiction, the other satiation. For example, food provides both calories and vitamins; the former is addictive, the latter counters the addiction. Note also that the rational habits tradition insists on a univariate process for memory loss: the habit stock depreciates at a constant rate. This convention can be attributed to the influence of growth theory on the theory of habit formation. Note, however, that while it is reasonable to assume a constant rate of depreciation for a physical capital stock, the habit stock is psychological and its depletion can also depend on other consumer activities. This is where the theory of # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 435 myopic addiction makes an important contribution. It alludes to a model where memory loss is a bivariate or multivariate process. Thus, we may consider a model in which non-durable consumption is habitforming and contractual saving, bonus income, or mortgage credit contribute to the depletion of the habit stock. To illustrate, suppose that the consumer is a new home owner with no other contractual commitments and earns no bonus income. As a by-product, we would expect the consumer's habitual behaviour to be modified. More precisely, habit persistence and consumer inertia will be moderated and the consumer will tend to behave as predicted by the LC=PIH. Here, the single objective will be to explore the implications of home ownership for habit formation effect. In this approach, the consumer will be a single, rational unit and the criticism directed at Shefrin and Thaler (1998) will not apply. Further, the model will allow for consumer heterogeneity since those with no experience in mortgage credit will be unable to change their habits and will, thus, fail to behave in accord with the LC=PIH model. Both nonseparabilities and a bivariate process for memory loss suggest habit modification in consumption.33 In a recent paper, Messinis (1997) explores two alternative models of habit modification: (i) non-durables (services included) and the durable stock are complements, and (ii) new consumer debt depletes the habit stock for non-durables. Both models predict that consumption will be less sluggish and more responsive to news about permanent income. In the second model, credit plays a dual role; it facilitates consumption smoothing and counters consumer inertia. In comparison to the LC=PIH model and standard habit formation, this model predicts that it is optimal to consume more when young and less when old if the credit effect on memory loss is strong. The model is consistent with recent evidence in * * * Jappelli and Pagano (1989, 1994) who show that countries with relatively low household indebtedness exhibit a higher degree of `excess sensitivity', Bauoumi (1993), de Brouwer (1996) and Olekalns (1997) who cannot reject the LC=PIH against the `excess sensitivity' alternative, and Shea (1995) who observes an asymmetric consumption response to expected changes in income: the response to income declines was stronger than that associated with income increases. Finally, the model suggests a new approach in public policy. First, it points to beneficial externalities associated with credit availability since credit is a source of habit modification. This highlights the need for further innovation in financial markets to satisfy consumer demand for modern saving devices. Second, the model suggests that concerns about the effects of an aging population on savings may not be warranted in post-deregulation. This is because the growth in household credit and=or rising rates of home ownership can lead to higher savings by the elderly in the future, assuming that credit constraints are relaxed. Last, the need for habit modification calls for a new approach in government policy away from mandatory and distortionary saving schemes (e.g. compulsory pension plans # Blackwell Publishers Ltd. 1999 436 GEORGE MESSINIS and tax deductions on superannuation funds) towards policies which are supportive of modern, demand-driven devices of habit modification. 6. Conclusion Habit persistence is an intuitively appealing hypothesis since it can explain several of the puzzles evident in the consumption literature. This paper illustrates this quality of habit formation but also acknowledges its limitations. In particular, existing formulations do not always perform well in empirical studies. In search for new ideas towards better models, the paper reviews the theory of addiction. The theory divides into two schools of thought. The first perceives the consumer as a single unit who is fully aware of the effects of addiction on future utility. Addiction is the outcome of consumer choice and involves a substantial increase in consumption over time. This consumption path has been modelled in terms of (i) an unstable steady state leading either to abstention or a high-consumption steady state, and (ii) an underestimation of the costs of addiction. With respect to withdrawal from addiction, prices, external shocks and more accurate information play an integral part. According to the second school of thought, the consumer deals simultaneously with several objectives. These originate in autonomous centres of motivation some of which are myopic. Addiction arises as a result of inner-conflict. Although conceptual problems regarding decision making, the consumer is seen as able to control addiction. In sharp contrast to the first school, research here has highlighted external stimuli as a means to managing addiction. Contractual behaviour, in particular, is seen as an important avenue through which behaviour is modified. Finally, regarding future research, the paper takes an eclectic view. It highlights the need for new research towards rational habit modification. It is worth considering two alternative paths towards habit modification: nonseparabilities and=or multivariate processes for memory loss. Acknowledgements I am most grateful to Nilss Olekalns, Chris Worswick and Olan Henry for helpful comments on earlier drafts of this paper. I am solely responsible for any errors. Notes 1. This has been mainly reported in studies using pre-financial deregulation data. 2. Attanasio and Weber (1995) demonstrate that the LC=PIH can also be consistent with a hump-shaped consumption profile when preferences are conditioned on demographics. Yet, Attanasio et al. (1995) also show that the hump-shaped profile of household characteristics closely resembles that of income. From a broader perspective, this poses an identification problem: to discriminate between preference changes purely due to demographics and those due to habit effects when demographics act as proxies for the latter. # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 437 3. There has been little progress towards explaning the disparity in wealth accumulation. The literature has focused on the lack of dissaving. This is often attributed to bequests seen as: (a) altruist, (b) payments for nonmarket services (e.g. care, `attention' or insurance), or (c) accidental due to an imperfect annuities market and uncertainty about lifetime and health. However, Hayashi (1997) has rejected the view that the extended family is altruistically linked. Hurd (1987, 1994) presents strong evidence in favour of accidental bequests but in Germany, where adequate insurance is provided by the social security system, the elderly never dissave (BoÈrsch-Supan 1992). 4. Hayashi (1997), however, advises that evidence of `excess sensitivity' alone does not necessarily signify the rejection of Hall's (1978) model. This is because an implicit liquidity constraint is built into the model by imposing a nonnegativity constraint on future consumption. 5. Campbell and Deaton (1989) and Flavin (1993) also attribute these two puzzles to a single cause but they emphasise consumer myopia or Keynesian-type behaviour. 6. This is often defined as the natural logarithm of the marginal rate of substitution when consumption is at a constant path. In a two-period intertemporal model it is the slope of the indifference curve at the point of intersection with a 45 ray from the origin. 7. See Olson and Bailey (1981). Yet, scepticism regarding the sign of  remains. Loewenstein (1987) and Price (1993) make a case for negative time preferences. 8. Also, there is a technical reason for rejecting the assumption of  0 (
) < 0: decreasing impatience implies a continuous increase in optimal saving which, in turn, leads to non-convergence in the distribution of wealth (Obstfeld 1990, pp. 51± 52 and Epstein 1987, pp. 73± 74). 9. This is in sharp contrast to standard models of time preference in which all wealth is eventually controlled by the most patient individuals. 10. The shadow price includes the net benefit of an additional unit of C implied by the habit stock in the future as well as the current market price of C. 11. A cubic utility function is implied here. 12. The BM model can also explain `binges' Ð prolonged oscillations between addiction and abstinence. This implies a technology with two stock variables; one stock being complement to current consumption, the other being a substitute. Dockner and Feichtinger (1993) and Feichtinger (1993) show that such a model leads to periodic or chaotic consumption cycles. 13. Consumers are fully aware of the future effects of addiction. Chaloupka (1991) succinctly shows that current consumption will positively correlate with past and future prices and consumption. Evidence seems to support this prediction (e.g. Chaloupka 1991 and Olekalns and Bardsley 1996). 14. Terminations of addiction with `cold-turkey' are attributed to discontinuities in the C ÿ S schedule due to joint convexity in C and S arising from severe degrees of addictions. 15. An Australian survey shows that 24% of ex-smokers had tried more than three times before they were able to quit (Australian Medical Association 1995). For America, Schelling (1984b) cites evidence showing that about 50% of smokers had tried to quit more than twice while 36% had made an unsuccessful attempt. 16. That is, the shadow price of S at the expected time of death is zero. 17. P is the mean of a sampling distribution of the true probability of being a nonaddict. 18. Consumers are misled into thinking that they are on the optimal path associated with certain nonaddicts and, in the process, are drawn into a steady state of addiction, given that S > Sc. # Blackwell Publishers Ltd. 1999 438 GEORGE MESSINIS 19. Winston gives no reason for assuming that only for addictive goods UM < UM . 20. When MZ < MRSMZ , there is no conflict and the model reduces to that of Stigler and Becker (1977). 21. Note, however, that conflict does not always imply that the individual is worse off when the long-run preferences are ignored. Winston clearly shows that with nonseparable utility in Z and past values of M, the individual may be better off, in the long-run, to `flip-flop' between U and U rather than abstain from M altogether. 22. Here, Taylor decomposes the Wundt effect discussed by Scitovsky (1992). This is an inverted U-shaped curve which depicts the sensation derived from the application of a stimulus. The novelty of the sensation initially produces `pleasantness' which positively relates to the intensity of the stimulus. Soon, however, familiarity with the sensation leads to satiation and `unpleasantness'. 23. Needs are ranked in terms of their relative importance to the consumer with physiological needs being of outmost importance. 24. Luxuries goods are linked to consumption activities requiring an unusually high level of skill by the consumer. Skilful activities, according to Scitovsky (1992) once again, are creative and, thus, produce a lower level of discomfort. 25. Lack of imagination and abstraction, and uncertainty about lifetime are the other two causes. 26. Indeed, the consensus in the literature is that the bonus system can partly explain differences between high-savers as in Japan, Italy, Korea, Singapore and Taiwan and low-savers as in the USA (Horioka 1990). 27. The view that homeowners are higher savers than renters is consistent with evidence in the USA, Canada (Bosworth et al. 1991) and Japan (Horioka 1988). Saving programs towards home ownership also stimulate total household=personal savings, see Engelhardt (1996) for Canada and BoÈrsch-Supan and Stahl (1991) for Germany. 28. See footnote 16 in Shefrin and Thaler (1988) which was deleted in Shefrin and Thaler (1992). 29. Note also that Ulysses' choice provides a clue to the question regarding the incentive payment in his principal-agent problem. Ulysses knows that the decision to precommit can only be implemented with the consent of his `doer' self. In order to entice the `doer' to approve of the rational decision to precommit, the `planner' offers him a contract which includes some instant gratification: the joy of majestic sounds. This is the substitution effect in Hoch and Loewenstein (1991) which says that strategies with an immediate reward can be highly effective in managing addiction. 30. This can be attributed to liquidity constraints (Miles 1992). However, Shefrin and Thaler (1992) dispute this on the basis that there has been little demand for `reverse mortgages' (loans advanced to elderly borrowers on the expectation that the debt will be repaid at the time of death). 31. There is some work in progress which attempts to incorporate game theory to make these models operational (Boymal 1997). 32. In the emerging literature of health psychology, evidence shows that exercising is an activity which assists smokers to quit and avoid relapse, heavy drinkers to reduce alcohol consumption and obese people to maintain weight loss (Green and Shellenberger 1991; chs. 11± 13). Activities such as exercising, listening to music, and hobbies play an integral part in what Taylor (1995, ch. 4) calls `lifestyle reballancing', i.e., sustained recovery from addiction. 33. Although as a concept it has a long history in philosophy, habit modification is more often associated with psychology. The school of `behaviorists', in particular, takes the # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 439 view that new habits can be formed in response to environmental stimuli via positive or negative reinforcements (Nye 1992). References Abel, A. B. (1990) Asset prices under habit formation and catching up with the Joneses. American Economic Review, Papers and Proceedings, 80, 38± 42. Akerlof, G. (1991) Procrastination and obedience. American Economic Review, Papers and Proceedings, 81, 2, 1 ± 19. Alessie, R. and Lusardi, A. (1997) Consumption, saving and habit formation. Economics Letters, 55, 103 ± 8. Archibald, R. B. (1994a) How many paychecks? an example of a self-imposed constraint. Economic Inquiry, 32, 4, 696 ± 702. Archibald, R. B. (1994b) Intertemporal trades and problems of self-control. Journal of Economic Behavior and Organization, 25, 1, 73± 91. Attanasio, O. P. and Weber, G. (1995) Is consumption growth consistent with intertemporal optimization? evidence from the consumer expenditure survey. Journal of Political Economy, 103, 6, 1119± 1157. Attanasio, O. P., Banks, J., Meghir, C. and Weber, G. (1995) Hump and bumps in lifetime consumption. National Bureau of Economic Research, Working Paper 5350, (November). Australian Medical Association (1995) Family doctor week survey. Australian Medical Association Report prepared by The Roy Morgan Research Centre, ACT, (July). Bayoumi, T. A. (1993) Financial deregulation and consumption in the United Kingdom. Review of Economies and Statistics, 75, 3, 536 ± 39. Bayoumi, T. A. and Koujianou, P. (1990) Consumption, liquidity constraints and financial deregulation. Greek Economic Review, 12, 2, 195 ±210. Becker, G. S. (1992) Habits, addictions and traditions. Kyklos, 45, 3, 327 ±45. Becker, G. S. and Murphy, K. M. (1988) A theory of rational addiction. Journal of Political Economy, 96, 4, 675 ± 700. BoÈhm-Bawerk, E. V. (1889) Capital and Interest. South Holland, IL: Libertarian Press, 1970. BoÈrsch-Supan, A. (1992) Saving and consumption patterns of the elderly: the German case, Journal of Population Economics, 5, 4, 289 ± 303. BoÈrsch-Supan, A. and Stahl, K. (1991) Do savings programs dedicated to home-ownership increase personal savings? Journal of Public Economics, 44, 265 ± 297. Bosworth, B., Burtless, G. and Sabelhaus, J. (1991) The decline in saving: evidence from house hold surveys. Brookings Papers on Economic Activity, 0, 1, 183 ± 241. Boymal, J. (1997) Alcohol addiction an intrapersonal coordination failure. Monash University, Economics Department, PdD Conference in Economics and Business, Papers and Proceeding, 5 ± 7 November, Perth, Australia. Braun, P. A., Constantinides, G. M. and Ferson, W. E. (1993) Time nonseparability in aggregate consumption: international evidence. European Economic Review, 37, 5, 897 ± 920. Browning, M. and Lusardi, A. (1996) Household saving: micro theories and micro facts. Journal of Economic Literature, 34, 4, 1797± 1855. Campbell, J. Y. (1996) Consumption and the stock market: interpreting international experience. National Bureau of Economic Research Working Paper 5610, (June). Campbell, J. Y. and Cochrane, J. H. (1995) By force of habit: a consumption-based explanation of aggregate stock market behavior. National Bureau of Economic Research, Working Paper 4995. Campbell, J. Y. and Deaton, A. (1989). Why is consumption so smooth? Review of Economic Studies, 56, 3, July, 357 ± 374. # Blackwell Publishers Ltd. 1999 440 GEORGE MESSINIS Carroll, C. D. and Weil, D. N. (1994) Saving and growth: a reinterpretation. CarnegieRochester Conference Series on Public Policy, 40, 0, 133 ± 192. Chaloupka, F. (1991) Rational addictive behavior and cigarette smoking. Journal of Political Economy, 99, 4, 722 ± 42. Christiano, L. (1989) Understanding Japan's saving rate: the reconstruction hypothesis. Federal Reserve Bank of Minneapolis Quarterly Review, 13, 2, (Spring), 10± 25. Deaton, A. (1992) Understanding Consumption. Oxford: Clarendon Press. de Brouwer, G. (1996) Consumption and liquidity constraints in Australia and East Asia: does financial integration matter? Economic Research Department, Reserve Bank of Australia, Macroeconomic Workshop, University of Melbourne 10± 11 (April). Dockner, E. J. and Feichtinger, G. (1993) Cyclical consumption patterns and rational addiction. American Economic Review, 83, 1, 258 ± 63. Dynan, D. E. (1993) How prudent are consumers? Journal of Political Economy, 101, 6, 1104 13. Earl, P. E. (1995) Microeconomics for Business and Marketing: Lectures, Cases and Worked Essays. Edward Elgar. Elster, J. (1979), Ulysses and the Sirens: Studies in rationality and irrationality, Cambridge University Press, Cambridge: UK. Engelhardt, G. V. (1996) Tax subsidies and household saving: evidence from Canada. The Quarterly Journal of Economics, 111, 4, 1237 ±68. Epstein, L. G. (1987) A simple dynamic general equilibrium model. Journal of Economic Theory, 41, 1, 68± 95. Ermini, L. (1997) A discrete-time consumption model under durability of goods, habit for mation and temporal aggregation. University of Hawaii and Osaka School of International Public Policy, manuscript (January). Fauvel, Y. and Samson, L. (1991) Intertemporal substitution and durable goods: an empirical analysis. Canadian Journal of Economics, 23, 1, 192 ± 205. Feichtinger, G. (1993) Strange addictive behaviour: periodic and chaotic binges. In W. E. Diewert and K. Spremann (eds.), Mathematical Modelling in Economics. Berlin: Stringer-Verlag. Feichtinger, G., Hommes, C. and Milik, A. (1997) Chaotic consumption patterns in a simple 2-D addiction model. Economic Theory, 10, 1, 147 ±73. Ferguson, B. (1996) A note on the interpretation of the rational addiction model. University of Guelph, Department of Economics, Discussion Paper 1996± 1, (January). Ferson, W. E. and Constantinides, G. M. (1991) Habit persistence and durability in aggregate consumption. Journal of Financial Economics, 29, 2, 199 ± 240. Fisher, I. (1930) The Theory of Interest. New York: Macmillan. Flavin, M. (1993) The excess smoothness of consumption: identification and interpretation. Review of Economic Studies, 60, 3, 651 ± 666. Flavin, M. (1985) Excess sensitivity of consumption to current income: liquidity constraints or myopia? Canadian Journal of Economics, 18, 1, 117 ± 136. Frank, R. H. and Hutchens, R. M. (1993) Wages, seniority, and the demand for rising consumption profiles, Journal of Economic Behaviour and Organization, 21, 3, 251 ± 76. Georgescu-Roegen, N. (1954) Choice, expectations, and measurement. Quarterly Journal of Economics, 68, 3, 503ÿ44. Goodfriend, M. (1992) Information-aggregation bias. American Economic Review, 83, 3, 508 ± 19. Green, J. and Shellenberger, R. (1991) The Dynamics of Health & Wellness: A Biopsychosocial Approach. Fort Worth: Holt, Rinehart and Winston. Hall, R. E. (1978) Stochastic implications of the life cycle-permanent income hypothesis: theory and evidence. Journal of Political Economy, 86, 6, 971 ± 87. Hayashi, H. (1997) Understanding Saving: Evidence from the United States and Japan. Cambridge: MIT Press. # Blackwell Publishers Ltd. 1999 HABIT FORMATION AND THE THEORY OF ADDICTION 441 Heaton, J. (1995) An empirical investigation of asset pricing with temporally dependent preference specifications. Econometrica, 36, 3, 681 ± 717. Hindy, A., Huang, C. and Zhu, S. H. (1997) Optimal consumption and portfolio rules with durability and habit formation. Journal of Economic Dynamics and Control, 21 (2 ±3), 50. Hoch, S. J. and Loewenstein, G. F. (1991) Time-inconsistent preferences and consumer self control. Journal of Consumer Research, 17, 4, 492 ±507. Horioka, C. Y. (1990) Why is Japan's household saving rate so high? a literature survey. Journal of the Japanese and International Economies, 4, 1, 49 ± 92. Horioka, C. Y. (1988) Saving for housing purchase in Japan. Journal of the Japanese and International Economies, 2, 3, 351 ± 384. Hubbard, R. G., Skinner, J. and Zeldes, S. P. (1994) Expanding the life-cycle model: precautionary saving and public policy. American Economic Review Papers and Proceedings, 84, 2, 174 ± 179. Hurd, M. D. (1994) Measuring the bequest motive: the effect of children on saving by the elderly in the United States. In T. Tachibanaki (ed.), Savings and Bequests. Ann Arbor: University of Michigan Press. Hurd, M. D. (1987) Savings of the elderly and desired bequests. American Economic Review, 77, 3, 298 ± 312. Iannacone, L. R. (1986) Addiction and satiation. Economics Letters, 21, 1, 95± 99. Jappelli, T. and Pagano, M. (1989) Consumption and capital market imperfections: an international comparison. American Economic Review, 79, 5, 1088± 1105. Kotlikoff, L. J. (1989) Intergenerational transfers and savings. In L. J. Kotlikoff (ed.) What Determines Savings?, Cambridge Mass.: MIT Press. Kotlikoff, L. J. (1987) Intergenerational transfers and savings. NBER, Working Paper 2237, (May). Loewenstein, G. (1987) Anticipation and the valuation of delayed consumption. The Economic Journal, 97, 666 ± 684. Mankiw, N. G. (1982) Hall's consumption hypothesis and durable goods. Journal of Monetary Economics, 10, 3, 417 ± 25. Messinis, G. (1997) Habit modification in consumption. The University of Melbourne, Economics Department, manuscript, (October). Miles, D. (1992) Housing markets, consumption and financial liberation in the major economies. European Economic Review, 36, 5, 1093± 1136. Mill, J. S. (1863) Utilitarianism. London: Dent & Sons, 1964. Mill, J. S. (1859) On Liberty. London: Dent & Sons, 1964. Muellbauer, J. (1996) Consumer Expenditure. In T. Jenkinson (ed.) Readings in Macroeconomics, Oxford (NY): Oxford University Press. Muellbauer, J. (1988) Habits, rationality and myopia in the life cycle consumption function. Annales D'Economie et de Statistique, 0, 9, 47 ± 70. Nye, R. D. (1992) Three Psychologies: Perspectives from Freud, Skinner, and Rogers. California: Brooks=Cole. Obstfeld, M. (1990) Intertemporal dependence, impatience and dynamics. Journal of Monetary Economics, 26, 1, 45± 75. Olekalns, N. (1997) Has financial deregulation revived the permanent income=life cycle hypothesis? The University of Melbourne. Department of Economics, Research Paper 564, (April). Olekalns, N. and Bardsley, P. (1996) Rational addiction to caffeine: an analysis of coffee consumption. Journal of Political Economy, 104, 5, 1100± 04. Olson, M. and Bailey, M. J. (1981) Positive time preference. Journal of Political Economy, 89, 1, 1 ± 25. Orphanides, A. and Zervos, D. (1995) Rational addiction with learning and regret. Journal of Political Economy, 103, 4, 739 ± 58. # Blackwell Publishers Ltd. 1999 442 GEORGE MESSINIS Osborn, D. R. (1988) Seasonality and habit persistence in a life cycle model of consumption. Journal of Applied Econometrics, 3, 4, 255 ± 266. Patterson, K. D. (1994) Consumers' expenditure on non-durables and services and housing equity withdrawal in the United Kingdom. The Manchester School, 62, 3, 251 ± 74. Pemberton, J. (1995) Income catastrophes and precautionary saving. University of Reading, Discussion Papers in Quantitative Economics and Computing, Series E Vol. III, 28, (February). Pischke, J. S. (1991) Individual income, incomplete information, and aggregate consumption. Princeton University, Industrial Relations Section, Working Paper 289, (August). Price, C. (1993) Time, Discounting and Value. Oxford UK: Blackwell. Runkle, D. E. (1991) Liquidity constraints and the permanent-income hypothesis: evidence from panel data. Journal of Monetary Economics, 27, 1, 73± 98. Scitovky, T. (1992) The Joyless Economy: The Psychology of Human Satisfaction (Revised Edition). New York: Oxford Uni. Press. Shea, J. (1995) Myopia, liquidity constraints, and aggregate consumption: a simple test. Journal of Money Credit and Banking, 27, 3, 798 ± 805. Shefrin, H. M. and Thaler, R. H. (1992) Mental accounting, saving and self-control. In G. Loewenstein and J. Elster (eds), Choice Over Time. New York: Russell Sage Foundation. Shefrin, H. M. and Thaler, R. H. (1988) The behavioral life cycle hypothesis. Economic Inquiry, 26, 4, 609 ± 43. Shelling, T. C. (1984a) Choice and Consequence. Cambridge, Mass: Harvard University Press. Shelling, T. C. (1984b) Self-command in practice, in policy, and in a theory of rational choice. American Economic Review, Papers and Procedures, 74 (2) 1 ± 11. Shi, S. and Epstein, L. G. (1993) Habits and time preference. International Economic Review, 34, 1, 61± 84. Startz, R. (1989) The stochastic behavior of durable and nondurable consumption. The Review of Economics and Statistics, 71, 2, 356 ± 63. Stigler, G. J. and Becker, G. S. (1977) De gustibus non est disputandum. American Economic Review, 67, 2, 76± 90. Tachibanaki, T (1994) Introduction: savings and bequests. In T. Tachibanaki (ed.) Savings and Bequests. Ann Arbor: University of Michigan Press. Taylor, L. D. (1992) Brain structure and consumption dynamics. In L. Phlips and L. D. Taylor (eds.), Aggregation, Consumption and Trade. Dordrecht: Kluwer. Taylor, S. E. (1995) Health Psychology (3rd ed.). NY: McGraw-Hill. Thaler, R. H. (1994) Psychology and savings policies. American Economic Review Papers and Proceedings, 84, 2, 186 ± 192. Thaler, R. H. (1992) How to get real people to save. In M. H. Kosters (ed.) Personal Saving, Consumption, and Tax Policy. Washington: AEI Press. Uzawa, H. (1968) Time preference, the consumption function, and optimum asset holdings. In J. N. Wolfe (ed.), Value, Capital, and Growth: Papers in Honour of Sir John Hicks. Chicago: Aldine. Venti, S. F. and Wise, D. A. (1989) But they don't want to reduce housing equity. NBER Working Paper 2859, (February). Weil, D. N. (1994) The saving of the elderly in micro and macro data. Quarterly Journal of Economics, 109, 1, 55± 81. Winston, G. C. (1980) Addiction and backsliding: a theory of compulsive consumption. Journal of Economic Behavior & Organization, 1, 295 ± 324. # Blackwell Publishers Ltd. 1999