Social Comparison and Wealth Inequality in a Leveraged Asset Market

Abstract

We hypothesize that upward social comparison of asset holdings among traders exacerbates leveraged asset bubbles because traders shift their frame of reference from profit maximization toward quantity maximization, increasing price momentum. In addition, asset prices should inflate even more in markets with wealth inequality because the relative reference shift becomes stronger. We test this theory within the standard asset market experiment environment, where we introduce the ability to borrow using leverage and treatments encouraging social comparison and manipulation of wealth inequality. We find that social comparison leads to asset overpricing, and its impact is the greatest when combined with wealth inequality. On the other hand, wealth inequality alone does not lead to greater asset price bubbles. These findings are consistent with housing market patterns prior to the financial crisis.

Keywords:

• Wealth inequality
• Social comparison
• Asset market experiment
• Peer effects
• Asset bubbles

Notes

1 A savings glut theory asserts that greater wealth inequality induces an endogenous increase in liquidity because the savings glut among the wealthy induces a response by the political or financial system to facilitate credit transfers (Rajan 2010; Kumhof, Rancière, and Winant 2015; DeMarzo, Kaniel, and Kremer 2008), which allows budget-constrained firms and households to purchase debt-financed assets and consumer goods, increasing economic growth (Cynamon and Fazzari 2016). Indeed, during the housing boom, financial intermediaries generated ample credit supply through large-scale subprime originations and securitizations (Mian and Sufi 2009; Nadauld and Sherlund 2013), and households took advantage by borrowing heavily to finance consumption and purchase illiquid assets (Bhutta and Keys 2016).

2 A limitation of our research design is that it does not identify subjects’ motivations for the reference shift. Emotions, however, have been shown in experimental settings to influence asset bubbles. Andrade, Odean, and Lin (2016) find that excitement among traders is associated with larger asset bubbles. Experiments by Schoenberg and Haruvy (2012) also influence trader emotions through upward and downward reference treatments where traders are informed of the maximum and minimum holdings in the markets.

3 Allowing the subjects to use prior dividend payments to buy shares means the cash to asset ratio is increasing over the 15 periods of the market. This dynamic has been found to foster bubble formation (see Kirchler, Huber, and Stöckl 2012). An alternative design would be to keep dividends in a separate account that could not be used for trading; however, such an approach lacks realism, as investors are free to use their own money as they see fit. Kirchler, Huber, and Stöckl (2012) also argue the declining nature of fundamental value in SSW markets encourages bubble formation due to confusion, but Cheung, Hedegaard, and Palan (2014) calls those claims into question.

4 Hence, traders in unequal sessions know that wealth inequality exists, but not the degree of inequality to avoid discouragement of poor traders.

5 In the instructions for treatments with an explicit social comparison, subjects were informed that they would be shown the maximum number of shares held by any trader. As described above, the subjects were not informed of the initial endowment when there was inequality. Therefore, the maximum number of shares held at the start of the session was withheld to avoid issues related to asymmetric information as that information would have revealed the magnitude of the inequality to poor subjects in the unequal treatment with social comparison, but not to the rich traders in that treatment nor to the rich and poor traders in the treatment with unequal wealth and social comparison absent.

6 While experimenter demand effects are concerning in many experiments, as argued by Zizzo (2010), in cases such as this where the experimenter demand effect serves to enhance the phenomenon being studied, the effect enhances external validity. In this case, the experimenter demand effect would be to encourage people to act as if they want to hold more assets, exactly the social comparison the manipulation is designed to foster.

7 No poor-type trader held the most shares after the first or third period; therefore, no poor-type trader was recognized by the experimenter.

8 In markets with social comparison absent, knowledge by poor traders that they have low endowments could incentivize them to borrow to purchase shares, which may increase asset bubbles in unequal markets relative to equal markets. This effect, however, is likely to be small. Indeed, we find no statistically significant difference in bubble metrics between equal and unequal markets when social comparison is absent.

9 As robustness, we use only accepted bids and we also compare bids in each of the 15 periods separately. In both cases, the results are similar. We also note that comparing bids across time periods can be misleading given the declining nature of fundamental value. For example, a specific bid amount placed in an early period may be well below fundamental value, but well above it in a later period. While one could normalize bids relative to fundamental value, we do not to be consistent with the literature. However, our main focus in comparing behavior across treatments at given points in the experiment and thus normalizing by fundamental value would not change our conclusions.

10 The noisy estimation is not surprising given the collinearity among the variables and the random differences in price dynamics across each of the 12 sessions. Regressions for poor-type traders had the best fit with an average of 93 observations and an average R² of 0.47. The same numbers for rich-type traders were 51 and 0.46, and for poor-type traders, 125 and 0.20.

11 In unreported tests, we run OLS regressions at the session level of bubble metrics on indicator variables for endowment, social comparison, and their interaction. The regressions affirm the results from the comparison tests.

12 We used the following procedure to generate simulated bids. First, for each period and bidder characteristic we identified the following percentiles of bids from the observed data: 0, 10, 25, 50, 75, 90, 100. We then constructed a cumulative distribution function that matched the empirical data at these points and linear interpolation between these points. We then drew a random number between 0 and 1 and used the inverse of the constructed cumulative distribution function to determine the simulated bid. A similar process was used to generate the simulated asks. We also conducted separate simulations assuming that bids (asks) followed a normal distribution with the same first and second moment as the observed distribution of bids (asks). The results are qualitatively similar between the two processes, but using the percentiles reduces effects from extreme outliers.

13 A similar result occurs when risk attitudes are based on the survey measure instead. The correlation between the two measures is 0.17. Holt and Laury (2002) show that incentivized methods of eliciting risk preferences are more accurate than hypothetical methods.

14 We are unable to assess the effects from cognitive reflection because the low number of bids by traders with high cognitive reflection in the sessions absent the social comparison effect was insufficient to generate reliable bid distributions.

Additional information

Funding

This work was supported by the Bank of America Research Fund Honoring James H. Penick, and the Arkansas Bankers Association Endowed Chair in Banking.