The Excess Smoothness Puzzle – An Experimental Investigation of Consumption Marcus Giamattei Johann Graf Lambsdorff Abstract We run an experiment where groups of six subjects determine their level of consumption with a target-level that is proportional to income. All known reasons for the existence of smoothed levels of consumption have been abandoned, such as financial reserves or borrowing at the aggregate level, habit persistence, sluggish arrival of information, current income deviating from permanent income or selfcontrol problems. Nonetheless our consumption varies less than its equilibrium prediction. While mathematically simple, the challenge of our game arises from, first, income being determined endogenously, second, subjects failing to observe that savings are not scarce at the aggregate level, third, equilibrium being driven by nonstationary investments and, fourth, depending on treatment the payoff-dominant solution might not be a dominant-strategy equilibrium. We find theoretical reasons for this behavior but suspect that our findings may also relate to sticky reasoning, the sluggish convergence towards equilibrium due to subjects being cognitively (rather than habitually) anchored by past data. JEL Classification: E20, E50, E12 Keywords: Beauty contest, absolute consumption hypothesis, permanent income hypothesis Marcus Giamattei is Research Assistants at the University of Passau, Germany, [email protected]. Johann Graf Lambsdorff is chair in economic theory at the University of Passau, [email protected]. Contact address: Innstrasse 27, D-94032 Passau. The authors are grateful to Lisa Einhaus for helpful comments. 1 1 Introduction In his General Theory Keynes (1936: 96-97) stated reasons for consumption to vary less than income, thus revealing excess smoothness relative to current income: “…men are disposed, as a rule and on the average, to increase their consumption as their income increases, but not by as much as the increase in their income… a decline in income due to a decline in the level of employment, if it goes far, may even cause consumption to exceed income not only by some individuals and institutions using up the financial reserves which they have accumulated in better times, but also by the government, which will be liable, willingly or unwillingly, to run into a budgetary deficit or will provide unemployment relief; for example, out of borrowed money.” Keynes relates this to government’s willingness to dampen fluctuations and to individual habits, standards of life that require time to adapt to new circumstances. This property has found widespread empirical support for the short term. Longer time horizons repeatedly revealed evidence in favor of a constant ratio between consumption and income, contrary to the Keynesian hypothesis. One proposal to explain the data has been proposed by Friedman (1956). He claims that the ratio is constant between consumption and permanent income. He relates short-term observations of excess smoothing to a statistical artifact: Households observe the ups and downs of current income but determine consumption proportionally to the expected, stable long-run level of their income. This implies that fluctuations of current income translate to less than proportional changes of permanent income and, thus, consumption. A less than proportional reaction of consumption to income is then a statistical illusion, because current income is an excessively volatile proxy for permanent income. The permanent income hypothesis has experienced various challenges. One criticism was raised by empirical studies which showed that consumers do not smooth out consumption as much as predicted by the permanent income hypothesis, (Hall and Mishkin 1982). There exists an excess reaction to permanent income, while consumption is smooth relative to current income. Flavin (1981) pointed to an excess dependence of consumption on past income. Campbell and Deaton (1989) and Deaton (1992) have thus advanced the term “excess smoothness puzzle”, the excess being in relation to current income. This puzzle has also become known as the Deaton paradox. Deaton (1992) suggests that habit persistence may explain a good deal of the puzzle. Habitual levels of consumption are seen to impact the marginal utility of consumption. A new steady state would not be approached immediately but rather smoothly as old habits fade out. For a review of theoretical and empirical approaches that include habits in the determination of consumption see Fuhrer (2000), Seckin (2001) and Attanasio (1999).1 Goodfriend (1992) observes that sluggish arrival and inattention to information can also account for explaining the puzzle. Aggregate shocks are imperfectly perceived such that consumption reacts only slowly to aggregate income. Deaton (1992: 171-174) observes that such lags in the arrival of information generate consumption patterns similar to those with habit persistence. Anomalies in the determination of consumption have recently been observed by reference to psychological elements (Shefrin and Thaler 1988) or neuroeconomics, (Brocas and Carillo 1 The life-cycle permanent income hypothesis has been criticized on further grounds, such as due to a bequest motive or market imperfections. These are not further discussed here as they contribute more to reasons why current income may retain some significance. Their contribution to the excess smoothing puzzle relative to current income is less clear. 2 2008). Consumers are seen to face a self-control problem. Discounting is steeper in the immediate future than in the further future. For a review of recent field experiments see DellaVigna (2009). This has been employed to depart from the permanent income hypothesis and restate the importance of current income, (Thaler 1990). This strand of research may also contribute to explaining the excess smoothing puzzle, (Akerlof 2007: 17). Households may impose discipline on themselves by following the rule that they need to work more if they want to consume more. Consumption would be tied to costly effort and subjects suffer from a psychological penalty when they finance consumption from current assets or future income (Shefrin and Thaler 1988). During a boom, households may obtain windfall profits. But these windfalls are effortless and little motivate households to increase consumption. A recession may go along with substantial efforts to resist deprivation but little income. These efforts may be seen as an entitlement to sustain a high level of consumption. But what might happen if all these reasons for smoothed consumption are absent? We conjecture that the puzzle might still survive. Smoothed consumption may also be caused by limited rationality. Subjects may fail in calculating the equilibrium or may expect others to fail. Finding equilibria can become cognitively demanding when income is unknown a-priori and must be forecasted from the behavior expected by all other subjects. This is the type of complexity already assumed by Keynes (1936: 44) when he notes for the level of income: “net income is what we suppose the ordinary man to reckon his available income to be when he is deciding how much to spend on current consumption.” The task of setting consumption thus requires first to forecast income. Forecast errors may then account for disequilibrium consumption levels. Such a forecast error was not considered by Keynes. He notes (1936: 122): “the logical theory of the multiplier [...] holds good continuously, without time-lag, at all moments of time” . As reviewed by Hoover (1997), Keynes failed in particular in presenting a consistent approach to the formation of expectations, treating short-term expectations as fulfilled or deviations to be negligible. Keynes assumed expected and realized income to be equal. He did not assume unplanned savings that might result from forecast errors. This was in contrast to adherents to the Stockholm school, who regarded income forecast errors to be important drivers of unplanned savings. Lindahl (1983: 235) criticizes Keynes from the perspective of the Stockholm school and describes how demanding the task of determining consumption can be: “If consumers plan to spend a certain fraction of their income during the period, but the income is determined only after the consumption purchases are finished, the only possibility of avoiding the distinction between the expected income, which is the basis of the consumption plans, and the realized income, which is the result of the carrying through of plans […] is to make them equal, i.e. implicitly to assume that individuals correctly anticipate their income.” In his lecture notes Keynes (1973: 181) deplored his approach: “I now feel that if I were writing the book again I should […] have a subsequent chapter showing the difference it makes when short-period expectations are disappointed.” This issue, unfortunately, has received little attention since. Extensive research was devoted to expectation formation, its capacity to explain short run deviations from long-run equilibria and whether policy can exploit these for its goals. But whether short-run equilibria are rationally determined in the Keynesian sense has, to the best of our knowledge, been less of a focus. A second difficulty for setting consumption arises from the question whether savings will suffice for the level of investment. If consumption is too high and savings fall short of 3 investment, costly borrowing must be arranged on imperfect markets. Likewise, excess savings that result from under-consumption must be transferred across imperfect markets to those in need of finance, producing similar frictional losses. The task to equilibrate savings and investments may be relevant for individual households that also engage in investments. If investors are separated from households, this task still arises at the regional level of an economy, if regions seek to balance their current account and capital markets that transfer savings from one region to another are imperfect. While this concern is relevant at the individual (or regional) level, a core contribution by Keynes (1936) has been that savings are never scarce at the aggregate level. In chapter 6, section II of the General Theory he argues: “Thus the act of investment in itself cannot help causing the residual or margin, which we call saving, to increase by a corresponding amount.” Chapter 14 is devoted completely to explain the equivalence of savings and investments and concludes: “the level of income must be the factor which brings the amount saved to equality with the amount invested.” Investments require additional financial means, but these are automatically created if banks hand out loans, investors use these means to make purchases and a multiplier process forces an increase in income that subsequently increases savings to an extent equal to the initial increase in investments. For a recent review of the debate see Lambsdorff (2011). We model an economy where this Keynesian assumption holds at aggregate level. But we hypothesize that subjects may fail to reason through this logic. The theoretically advanced will observe that sufficient aggregate savings are always generated, suggesting that frictional costs for excess or scarce savings can be disregarded. But others may fail to observe this logic. They might fear that increased consumption reduces their individual savings without recognizing positive feedback loops that arise from increased consumption by others and a multiplier process that raises income. Responding to increased investments they are thus tempted to reduce rather than increase consumption. This postpones convergence towards equilibrium. 2 Previous Experimental Evidence Laboratory macroeconomic experiments have gained prominence lately, as evidenced in the comprehensive survey by Duffy (2008). Quite a number of experiments have focused on commodity prices. The main challenge in these experiments is to learn to forecast correctly. These experiments have thus been labeled Learning to Forecast Experiments (LtFEs), (Hommes 2011). Subjects are not provided complete information on the quantitative structure of the underlying model. At the core of these experiments lies the question of whether equilibria can be approached by help of adaptive learning. For a more extensive review see Lambsdorff et al. (2011). Another branch of experimental research, which acts under limited information as well, deals with inflation forecasting in complex New Keynesian Dynamic Stochastic General Equilibrium (DSGE) Models, (Adam 2007; Pfajfar and Zakelj 2009; Assenza et al. 2011). Other experiments provide players with complete information and find evidence against monetary neutrality, (Fehr and Tyran 2001; 2008; Sutan and Willinger 2009). More complex models require subjects to determine prices and quantities simultaneously, (Noussair et al. 2011; Petersen 2011; Davis and Korenok 2011; Roos and Luhan 2010), considering different types of menu costs and their potential impact on quantities. A result that is close to our study is reported by Petersen (2011). She observes under-consumption by households, particularly when these have negative bank balances. This allows Petersen to trace under-consumption to debt-aversion. In reaction to decreased nominal interest rates 4 only slightly more than half of the subjects increased consumption, while almost a fifth even lowered it. She observes that “even with multiple repetitions subjects had considerable difficulty forming rational expectations.” Laboratory beauty contests have been employed to investigate the cognition of reasoning processes, and they reveal some similarity to our study. Nagel (1995) and Stahl and Wilson (1995) report the results from such a laboratory guessing game. In this experiment subjects are asked to pick a number between 0 and 100. The player whose number is closest to p (0<p<1) times the average of all numbers chosen wins a fixed prize while all other players earn nothing. The iterated elimination of (weakly) dominated strategies implies that only 0 survives as the equilibrium number. However, subjects substantially deviate from this equilibrium point. Average numbers are usually between 20 and 30 for p=2/3 and distributions of number choices show prominent spikes at 33 and 22. In order to explain these findings, both studies propose some boundedly rational refinements to the process of iterated application of dominance: First, level-0-players are defined to randomly select numbers between 0 and 100, the average value being 50. This value then serves as a focal point for more sophisticated players. Level-1 players best respond to level-0 players, thus choosing 33. Level-k players best respond to the assumption that all others are level-(k-1) players. With these adjustments, participants are found to obey two to three steps of iterated dominance rather than an infinite number (Nagel 1995; Ho, Camerer and Weigelt 1998). When being played repeatedly, in higher rounds level-0 players stop picking randomly, but employ the previous round’s average number as a starting value instead. This implies that repeated play generates convergence towards equilibrium. But this convergence can be particularly slow and strongly dependent on how the game is framed, (Ho, Camerer and Weigelt 1998: 950; Duffy and Nagel 1997: 1699). These findings, observed for interaction among subjects, reveal some similarities to heuristics that can be found at the individual psychological level. Tversky and Kahnemann (1974) observe how subjects start with acquainted pieces of information (anchor) when estimating unknown quantities. They then adjust the estimates until a satisfactory result is achieved. This adjustment is seen to remain incomplete as subjects test only whether the outcome is plausible but not whether it is perfect, (Epley and Gilovich 2006). We believe that subjects are limited in their capacity to find an equilibrium level of consumption, either because they have limited computational skills or because they do not belief that other subjects can calculate the equilibrium. For our experiment we prefer models with complete information. Departure from equilibrium in LtFEs may arise when the true model has not yet been detected and an efficient rule for forecasting has not yet been found. The same may be true for complex models where subjects need longer to learn. Our experiment is almost as simple as the above mentioned guessing games. But we let subjects determine consumption across many rounds and confront them with a nonstationary, exogenously given level of investments. To the best of our knowledge, while non-stationary data are standard to macroeconometrics, they represent a novelty in experimental macroeconomics. A non-stationary shock is cognitively demanding to subjects and allows us to observe how incentives to approach equilibrium are overshadowed by the arrival of news. While limits to rationality have already been widely observed in behavioral macroeconomics, it remains unclear whether they would disappear over time as subjects improve their understanding of the underlying model. We design a very simple model to test whether and when deviations from rationality can be observed in spite of its simplicity, this way not being caused by subjects’ failure to initially understand the given task. 5 3 Experimental Design We build on a very simple consumption model based on a Keynesian multiplier process with consumption , investment I and savings , where consumption together with investment determines income , which again determines consumption defined by with marginal and average propensity to consume being set to . Solving this simple model yields and . This implies automatically that and that total savings equal total investment. We implemented this model in the laboratory with the following design. An experimental economy consisted of 6 subjects who played rounds of the game. Each player was instructed to act as an adviser to a household in an economy consisting of 6 regions. His advisory task was to decide the level of consumption for the region’s single household in a range between and in every period. was the experimental currency used throughout the whole experiment. In each of the 6 regions a computerized investor determined levels of investments , identical across all regions. The value was announced prior to the consumption decision. While making the decision on consumption the household’s current level of income was unknown to the subjects. Income in each period was calculated as the sum over all consumption decisions and the total sum of investment divided by the number of players. That means that income was distributed equally across all households2 and given by ∑ ̂ , where ̂ . Subjects were incentivized to meet up to two different goals to maximize their payoff. First consumption should equal a target level on consumption ̅ , so that they aimed to minimize | ̅ |. This was explained by the fact that less consumptions does not satisfy the current lifestyle while excess consumption renders savings being short of their desired level. This goal implements the Keynesian multiplier process in our experimental economy. The second goal was to balance out savings and investment . This means |, which was motivated by the fact that cost similar to a use of the to minimize | capital market would occur. For the investor in a region it is easy to access the savings of the household of the same region , while organizing extra savings from other regions would be costly. The same holds true, if investments are higher than savings and additional savings have to be collected from other regions. As is known at least since Hanson (1949: 220), even in disequilibrium aggregate savings will necessarily be equal to aggregate investments. To see this, assume n subjects picking arbitrary values for consumption, with . ∑ Aggregate income then amounts to . But aggregate savings are determined by -∑ such that they must always equal investment. But they may differ in percapita-terms thereby causing costly deviations. Both deviations decreased the adviser’s payoff per round. Subjects were provided with an initial payoff at the beginning of all rounds and an endowment for every round . That means their overall payoff was given by ∑ | ̅| | | (1) 2 As we only used integer values a remaining part of income could not be distributed equally. Subjects thus took turns in getting one unit of the rest. 6 where is a weighting factor for sanctioning the savings deviation. The level of investment was given by a non-stationary random walk with and . We used a version of this time series, where an ADFtest, , produced insignificant coefficients and close to zero. These made sure that the process did not, by random selection, turn out to be stationary or characterized by serial correlation. We used the same sequence of in all treatments. The theoretical solution can be derived by maximizing (1) with respect to . Given that all subjects do the same calculus (i.e. ) it can be shown that the maximization ( ) as | | is always zero in equilibrium, since by problem is solved by construction and . This yields the payoff-dominant equilibrium of . Whether this is a dominant-strategy equilibrium is discussed below. Both the levels of investment as well as the equilibrium solution are displayed in figure 1. Figure 1 - Investment and Payoff-Dominant Equilibrium Level of Consumption But what about the adjustment process towards equilibrium? As in the beauty contest by Nagel (1995) this can be described by iterated elimination of (weakly) dominated strategies. Thereby player takes the average decision of all other player into account and maximizes (1) in order to calculate his best response. Inserting for consumption and income we obtain a minimization problem for both deviations: | ( | )| ( |( )| ) | | (2) | 7 Figure 2 - Reaction Function for Different for For equation (2) can be solved easily and yields (3). In figure 2 this reaction function is depicted with a red line for an investment level of . (3) Given this investment of and player assuming that all other players will choose their consumption level it is not optimal for m to choose , but to set consumption according to ( ) (see the dotted lines and arrows in figure 2). As player plays best response we label him level-1-player in contrast to all other players not doing best response and therefore being labeled level-0-player. Suppose now all other players – make the same calculus as player . It would then be optimal for player to set ( ) and act as a superior level-2-player in contrast to all other players being level-1. Now again all other players may act as level-2-players as well. Iterative deletion of strictly dominated strategies brings about the dominant-strategy Nash-equilibrium with . For the decision is more complicated. The first absolute term could be solved by the same calculus according to (3) as for , but the second absolute term incentivizes player to choose consumption as close as possible to the consumption of all others and | |. thereby minimizing This creates a trade-off between playing a response rule like (3) (and thereby ignoring the second deviation) versus “follow-the-crowd”. For the given set of consumption possibilities in the interval [0;100] with the second deviation is important enough so that best response is given by just setting . There is no incentive to deviate from the decisions of all other players. This reaction function is marked with a blue line in figure 2. As there are no incentives to deviate from other players’ 8 decisions a single player has no incentive to initiate a reasoning process towards equilibrium as described above. Disequilibrium values equally chosen by all subjects are not weakly dominated and iterated elimination of weakly dominated strategies will fail to converge strategy choices towards the payoff-dominant equilibrium. This equilibrium is thus not a dominant-strategy equilibrium. Still, the payoff-dominant equilibrium might be approached if enough subjects are capable of calculating it and believe others to do the same. With equation (2) implies that the first term should always be more important than the ⁄ , while the second term. Variations of by 1 imply changes in the first term of second term varies by ⁄ . Thus, even if a change in the second term counters the one of the first term, it never outbalances it. It then suffices to recognize the first term for minimizing deviations. The payoff-dominant solution is thus obtained by elimination of strictly dominated strategies. However, given that the game was only played with integers, the minor difference in changes between the two terms often does not materialize. We thus argue more carefully that elimination of weakly dominated strategies still brings about the payoff-dominant solution. We thus depict best response behavior by the green, shaded area in figure 2. This means that can also be seen as an inverse indicator on how strong the forces are that push subjects towards equilibrium. While indicates rational herding towards the equilibrium that affects all subjects, with some subjects in this area may feel comfortable with not moving towards the payoff-dominant solution (and thereby disregarding that they can make others better off). With there is no automatic driver at all but only the captaincy of single subjects may guide the group towards equilibrium. 4 Treatments We implemented three different treatments, see table 1. As treatment variable we changed the parameter in order to vary the importance of savings deviating from investment, S-I. In order to ensure a fair reimbursement of our subjects we increased the endowment with a rise in the parameter . In every treatment subjects played 24 rounds of the game, where the first four rounds were used for practicing purposes. Treatment Importance of S-I 0: Baseline 1: Normal Savings 2: Double Saving 0 1 2 None Normal Double Endowment round 10 20 22 per Rounds (practice/with payoff) 4/20 4/20 4/20 Table 1 - Treatment Overview In the baseline treatment subjects just had to focus to meet their consumption target, which was equal to 80% of income. With the fact that savings should match investment was irrelevant in this context. This design is similar to a beauty contest with an inner solution and can be solved by iterated elimination of dominated strategy as shown above. We expected that subjects may succumb to similar cognitive limitations as in the beauty contest and may fail to approach equilibrium. We then increased the cognitive challenge by letting subjects also consider a target for savings. With subjects were sanctioned for their savings deviating from investment. In order to avoid ordering effects we varied the order in which both deviations were 9 reported. This permutation revealed no significant differences, which may also act as an additional robustness check for our design. We hypothesize that adding this additional target may detract subjects from equilibrium. An exogenous increase in investments implies that savings must be increased. Savings are commonly increased at the individual level by reducing consumption. This logic may serve as a salient anchor to guide subjects’ behavior. They may observe that increased investment also increases income, such that a reduction of consumption is unnecessary. But few will be as sophisticated as to observe that aggregate savings are always generated to a sufficient amount, such that their task can be simplified to setting consumption to its target value. Due to this we hypothesize that increasing the penalty for missing the saving’s target may drive away the game from equilibrium. We thus implement a second treatment with . Under this regime we expected deviations of savings from investment were sanctioned very extensively. This should minimize any attempt by a single player to raise consumption to its target value because deviation from the others consumption level would mean large payoff losses due to the savings deviation. We think that subjects still may be capable of coordinating their decisions but may largely fail in approaching the payoff-dominant equilibrium. Still the theoretical solutions would predict that herd behavior should be larger under these circumstances as “follow-the-crowd” is optimal considering best response. That means that the only forces which push in the direction of the payoff-dominant equilibrium are players who understood the model. As their presence may vary we expected a larger variation in the third treatment. 5 Experimental Procedures The experiment was conducted computer-based at the Passau University Experimental Laboratory (PAULA) using z-Tree (Fischbacher 2007). Figure 3 –English Translation of one Instruction Screen – Baseline Treatment 10 Upon arrival, subjects were randomly seated in the laboratory and publicly instructed 3 about the purpose of the game, its expected length, dos and don'ts and about (standard) payment and blindness procedures. In order to increase overall understanding of the rules, the first screens explained the game in a detailed manner using a step-by-step approach that was found to be perceived as intuitively appealing by pilot subjects (see figure 3 for an example). Through the whole experiments a graph visualized the functionality of the economy and displayed the values of the previous period. Subjects were provided with a calculator. The first four rounds (1*, 2*, 3* and 4*) in each treatment were reserved for learning, thus payoffs in these rounds were hypothetical. Actual payoffs were achieved in the following 20 rounds. Each subject participated in only one treatment (between-subjects design). After the experiment subjects had to fill out a questionnaire on demographic variables. In all but the baseline treatment they were asked as well on how they evaluated the deviation of savings and target consumption for their decision process. Subjects should classify the importance of both savings and target consumption on a scale from 1 (not important at all) to 5 (very important). Unsurprisingly, a rise in induced subjects to increase their importance of savings. But despite high importance of savings some groups kept importance of target consumption on a high level. Throughout the entire experiment we provided feedback on all relevant information (see figure 4). At the end of the experiment, each subject received the sum of earned at an exchange rate of 1 Taler = 4 Eurocent. Total payoffs were rounded to 5 Eurocent. Payments were disbursed by a third person. Figure 4 – Decision Screen for Round 7 – English translation – Baseline Treatment 3 Instructions are available upon request. 11 6 Descriptive Results The experiment was conducted in seven sessions of 12 to 18 students from the University of Passau at February 9th 2012, 26th 2012 and March 6th. In total, 114 subjects participated and formed 19 autonomous laboratory economies. Total payoffs to the 114 participants amounted to 1186.20 €. Altogether 2/3 of the participants were female with an average age of 23.63. Participants were students of different field of studies, predominantly social sciences, business, economics and law, where only 15% had taken a lecture in macroeconomics. For payoffs and durations of all treatments see table 2. Baseline Treatment 1 Treatment 2 All Time (in min.) Payoff Number of economies Total Instructions Average Min4 Max Average per Hour 6 10 3 19 52.96 71.19 91.46 71.87 17.28 24.67 34.39 25.45 7.42 € 12.57 € 9.09 € 9.70 € 2.00 € 3.00 € 3.00 € 2.00 € 9.00 € 15.96 € 13.40 € 15.96 € 8.41 € 10.60 € 5.96 € 8.32 € Table 2 - Payoff and Time for Different Treatments A first grasp of the results is presented in figures 5-7. The first four rounds for learning are separated by a vertical line. The figures depict the median values of chosen consumption, alongside with the equilibrium prediction and upper and lower quintiles of individual choices. Figure 5 depicts values for the baseline treatment. As can be seen, while consumption is too high initially it over time converges to equilibrium. While the upper and lower quintiles denote some heterogeneity, subjects seem to react to investments according to theoretical predictions. Our hypothesis that subjects may fail to approach equilibrium, as they do in one-shot beauty contests or in Lambsdorff, Schubert and Giamattei (2011) in repeated, non-stationary play, was not confirmed. 4 The payoff was restricted downwards not to be lower than 2€ (baseline treatment) or 3€ (first and second treatment). Negative payoffs and payoff less than these amounts were rounded up to these minimum payoffs. 12 Figure 5 - Individual Consumption - Baseline Treatment Figure 6 - Individual Consumption - Treatment 1 Figure 6 depicts values for the first treatment where subjects are additionally confronted with targeting savings. As can be seen, consumption varies less than its equilibrium prediction. Only the upper and lower quintiles are partly capable of coming close to equilibrium values. Heterogeneity appears to be larger than in the baseline treatment. 13 Figure 7 - Individual Consumption - Treatment 2 Figure 7 depicts values for the second treatment where deviations from the savings target were more severely penalized. Subjects revealed severe limitations in identifying the equilibrium. One group was able to come close to equilibrium, being denoted by data for the upper quintile. The two other groups remained stuck in disequilibrium with strong underconsumption. Heterogeneity increased over time. 7 Regression Analysis We focus on average group levels of consumption ̂ and regress this variable on the announced level of investments: ̂ The baseline treatment shows that subjects were able to approach the equilibrium. The constant is close to zero. Thus, there is no autonomous consumption in line with our equilibrium prediction. Investments enter with a coefficient of 4.2, which is close to the equilibrium prediction of 4. 14 Method: Ordinary Least Squares and ADF-test.a) Dependent Average Group Consumption, Variable Independent Baseline 1. Treatment 2. Treatment Variable -2.68 23.2 18.6 1 Constant (-2.7) (19.5) (3.6) 4.2 2.17 1.31 2 Investment, it (52.4) (22.5) (3.1) Rounds 1-20 1-20 1-20 Total Obs. 120 200 60 2 R 0.96 0.72 0.14 Dependent Change in residuals from above regression, Variable t 1 Constant 0.45 (1.7) -0.33 (-4.97) 2 lagged residual, t 1 0.09 3 lagged change (0.9) of residual, t 1 Total Obs. 108 2 R 0.19 a) t-statistics in parenthesis. -0.03 (-0.1) -0.08 (-1.97) -0.15 (-0.3) 0.03 (0.93) -0.18 (-2.3) -0.02 (-0.1) 180 0.07 54 0.02 Table 3 - Time Series Regressions for Average Group Consumption A positive constant would depict autonomous consumption, even if the equilibrium predicts this term to be zero. We carry out an ADF-test on the residuals, t, of the type t 1 2 t 1 3 t 1 t and report the respective values. The critical McKinnon values for the coefficient 2 are -2.57 (10% error level) and -2.88 (5% error level). A note of caution is required. With only 20 observations per group convergence is not very strong and error levels may be measured with imprecision. In the baseline treatment the ADF-statistic reveals a t-statistic of the lagged residuals of 4.97, which denotes significance at the 1-percent level. The variables thus cointegrate. The first treatment reveals some considerable deviations from equilibrium. The constant signifies autonomous consumption and the impact of investment is too low. The ADF-test fails to be significant at the 10% level. Subjects thus maintained difficulties in finding equilibrium values and heterogeneity remained strong throughout the game. The second treatment reveals how strongly this confusion can be amplified. Investment still exerts a significant positive impact, but with a coefficient of 1.31 the magnitude is only about a third of the equilibrium prediction. With an R2 of 0.14 the explanatory power of the regression is poor and there is no cointegration of the variables. This mirrors our previous description of the data. For some groups we were able to crash the economy. 15 8 Conclusions and Outlook Our experiment revealed how limited reasoning may contribute to smoothed consumption. The endogeneity of income was manageable by subjects, bringing about payoff-dominant equilibrium levels of consumption. But an additional savings target turned out to be troubling to subjects. In line with theoretical predictions, equilibria can no longer be determined by iterated elimination of dominated strategies. But we hypothesize that also the task may have cognitively overburdened subjects. When investments must be financed by own savings, subjects may be tempted to reduce consumption in response to increased investments, failing to recognize the Keynesian logic that savings are never scarce at the aggregate level. While our findings thus cast a first light on limited reasoning in consumption decisions, much remains to be done. Are deviations from equilibrium well described by theory or are framing effects contributing to their emergence? As shown theoretically, rather than targeting savings an equivalent task can be framed by targeting average consumption. It is up to future research to identify whether this simpler task induces subjects to equilibrium play, even when theory suggests that the payoff-dominant equilibrium can no longer be approached by iterated deletion of dominated strategies. If not, this would suggest that the savings target involves non-trivial framing effects. Other treatments are equally promising and can provide extensions to our model. Would original sin, the asymmetric penalizing of insufficient savings but not excess savings, contribute to global underconsumption? Can this be remedied if a hegemon, such as the United States, is exempt from this restriction? Would overconsumption result if debtor nations are not fully responsible for their excessive consumption, for example due to their hopes to be bailed out? Our model provides a workhorse to tackle these issues for experimental subjects that face cognitive limitations. References Adam, K. (2007): Experimental Evidence on the Persistence of Output and Inflation. The Economic Journal 117 (520): 603–636. Akerlof, G. (2007), “The Missing Motivation in Macroeconomics,” The American Economic Review, Vol. 97 (1): 3-36 Assenza, T.; Heemeijer, P.; Hommes, C.; Massaro, D. (2011): Individual Expectations and Aggregate Macro Behavior. DNB Working Paper (298). Attanasio, O. (1999), “Consumption,” Handbook of Macroeconomics, Volume 1, ed. by J.B. Taylor and M. Woodford: Elsevier Science. Brocas, I. and J. Carrillo (2008), “The Brain as a Hierarchical Organization quick,” The American Economic Review, Vol. 98 (4): 1312-1346. Campbell, J. and A. Deaton (1989), “Why Is Consumption So Smooth?”, Review of Economic Studies, vol.56: 357-74. Davis, D.; Korenok, O. (2011), “Nominal price shocks in monopolistically competitive markets: An experimental analysis,” Unpublished manuscript, Virginia Commonwealth University. Deaton, A. (1992), Understanding Consumption, (Oxford: Clarendon Press) DellaVigna, St. (2009), “Psychology and Economics: Evidence from the Field,“ Journal of Economic Literature, Vol. 47(2): 315–372 16 Duffy, J. (2008): Macroeconomics: A Survey of Laboratory Research. In John Kagel, Alvin Roth (eds.): The Handbook of Experimental Economics (forthcoming). 2nd ed.: Princeton University Press. Available online at http://www.econ.pitt.edu/papers/John_hee11.pdf, checked on 27/09/2011. Duffy, J. and Nagel, R. (1997), “On the Robustness of Behaviour in Experimental Beauty Contest Games,” The Economic Journal 107 (445): 1684–1700. Epley, N.; Gilovich, T. (2006): The anchoring-and-adjustment heuristic. Psychological Science 17: 311-318. Fehr, E.; Tyran, J. (2001): Does Money Illusion Matter? American Economic Review 91 (5): 1239–1262. Fehr, E.; Tyran, J. (2008): Limited rationality and strategic interaction: the impact of the strategic environment on nominal inertia. Econometrica 76 (2): 353–394. Flavin, M. (1981), “The Adjustment of Consumption to Changing Expectations About Future Income,” Journal of Political Economy, vol.89: 974-1009. Fuhrer. J. (2000), “Habit Formation in Consumption and its Implications for MonetaryPolicy Models,” American Economic Review, Vol. 90(3): 367-390. Friedman, M. (1956), A Theory of the Consumption Function, (Princeton: Princeton University Press). Goodfriend, M. (1992), “Information-Aggregation Bias,” American Economic Review, Vol. 82 (3): 508-519 Hall, R. and F. Mishkin (1982), “The Sensitivity of Consumption to Transitory Income: Estimates from Panel Data on Households,” Econometrica, vol.50: .461-81. Hansen, A. (1949), Monetary Theory and Fiscal Policy. New York: McGraw-Hill. Ho, T.-H.; Camerer, C.; Weigelt, K. (1998): Iterated Dominance and Iterated Best Response in Experimental "p-Beauty Contests". The American Economic Review 88 (4): 947–969. Hommes, C. (2011): The heterogeneous expectations hypothesis: Some evidence from the lab. Journal of Economic Dynamics and Control 35 (1): 1–24. Hoover, K.D. (1997), “Is There a Place for Rational Expectations in Keynes’ General Theory?”, in Harcourt, G.C. and P.A. Riach (ed.), A "second edition" of The general theory, Vol. 1 Keynes, J.M. (1936), The General Theory of Employment, Interest, and Money. London, UK: Macmillan. Keynes, J.M. (1973): The Collected Writings of John Maynard Keynes. The General Theory and after. Part 2. Defence and Development. (Macmillan, London). Lambsdorff, J. Graf (2011), “Savings and Investments – an Old Debate in Times of Trouble,” Journal of Post Keynesian Economics, Vol. 33 (4): 645-666. Lambsdorff, J. Graf, M. Schubert and M. Giamattei (2011), “On the Role of Heuristics – Experimental Evidence on Inflation Dynamics,” Passauer Diskussionsbeiträge, Volkswirtschaftliche Reihe, V-63-11, University of Passau, Germany. Lindahl, E. (1983), “On Keynes’ Economic System. Part I-II.” John Maynard Keynes: Critical Assessments, ed. By J.C. Wood, (Croom Helm, London): 221-252. Nagel, R. (1995): Unraveling in Guessing Games: An Experimental Study. The American Economic Review 85 (5): 1313–1326. Noussair, C. N.; Pfajfar, D.; Zsiros, J. (2011): Frictions, Persistence, and Central Bank Policy in an Experimental Dynamic Stochastic General Equilibrium Economy. EBC Discussion Paper 2011-006. Petersen, L. (2011): Nonneutrality of Money, Preferences and Expectations in Laboratory New Keynesian Economies, unpublished manuscript, University of California Santa Cruz. Pfajfar, D.; Zakelj, B. (2009): Experimental Evidence on Inflation Expectation Formation. Tilburg University-Center for Economic Research Discussion Paper 2009 (7). 17 Roos, M.; Luhan, W. (2008), Are expectations formed by the anchoring-and-adjustment heuristic? An experimental investigation. Ruhr Economic Papers 54, Technische Universität Dortmund. Seckin, A. (2001), "Consumption-leisure choice with habit formation," Economics Letters, Elsevier, vol. 70(1): 115-120. Shefrin, H. and R. Thaler (1988). “The Behavioral Life-Cycle Hypothesis.” Economic Inquiry, Vol. 26(4): 609-43. Stahl, D.; Wilson, P. (1995): On Players Models of Other Players: Theory and Experimental Evidence. Games and Economic Behavior 10 (1): 218–254. Sutan, A.; Willinger, M. (2009): Guessing with negative feedback: An experiment. Journal of Economic Dynamics and Control 33 (5): 1123–1133. Thaler, R. H. 1990. “Anomalies. Saving, Fungibility, and Mental Accounts.” Journal of Economic Perspectives, Vol. 4(1): 193-205. Tversky, A.; Kahneman, D. (1974): Judgment under uncertainty: Heuristics and biases. Science 185 (4157): 1124. 18
© Copyright 2026 Paperzz