I think that this crisis has demonstrated that the cost of waiting to clean up asset bubbles after they burst can be very high. That suggests we should explore how to respond earlier.
William Dudley, New York Federal Reserve President (2009)

Bubbles are very serious phenomena that occur rather frequently, and are usually the cause of deep economic crisis. Furthermore, Bubbles are puzzling and pose questions about agents rationality and market efficiency (Stiglitz 1990). Policies regarding bubbles have often been reactive. In Quintero (2012), we suggest a preventive policy based on experimental results inspired by concepts drawn from behavioral economics, from mental accounting (Thaler, 1999) in particular. Mental accounting was developed using framing and prospect theory (Tversky and Kahneman 1981). It states that our minds assesses gains and losses in relation to a reference point, and classfies them in separate categories, like different items in a budget. This suggests that the reason people are more willing to bet again their casino gains than their pocket money is that they store these potential gains (or losses) in different accounts, and compare them to different reference points. Similarly, this could be the reason why sometimes investors are willing to keep long positions in assets that seem to be valued above their expected stream of dividends (or fundamental value) when they have experienced gains in the recent past. The proposed policy suggests that having casino winners cash their gains and take a walk outside, away from the exciting lights and beeps, can make cool off and reset their reference point, integrating the bet gains into their cash mental account and double think their decision of betting them again. Similarly, policies that change the reference point with which investors value their assets can make them at least partially more sensible to losses and risk, as they pool both financial gains in speculative and safer assets in the same mental account. Our proposed policy does not require prediction of the bubble and can be applied at any moment to bubble vulnerable markets.

Bubbles are relevant for us today

In an efficient market, asset prices should converge to the discounted present value of their dividend stream. In housing markets, this is equal to the housing services stream, or an equivalent stream of implied rent payments. In the absence of bubbles, prices should only change as a result of variations in the fundamentals of the asset (changes in the expected dividend or services stream) or of the economy. Deviations should not persist because they give way to arbitrage opportunities (Fama 1970). However, in practice these deviations last longer than predicted by theory (Tirole 1982).

Asset bubbles have played a starring role in most recent financial crises. They have damaging economic long term effects, having particularly damaging effects on low income households who lack diverse portfolios, as revealed by the latest Survey of Consumer Finances. Latin America is by no means exempt from these phenomena. Several bubbles have occurred here (Seyfried et al 2010), among which housing bubbles have had especial importance. Especially in this sector, the issue is very current. Since the end of the last decade, foreign investment has been flowing into the housing market in Latin America (Knowledge@Wharton Nov. 2010), most likely as a reaction to the downfall of the US market after the subprime crisis. Since 2011, the region's housing market is experiencing a boom that could call for caution in order to ensure there will not be a bubble bust (EC-IILS¹2010) as pointed out by several reports by the IMF (Cubeddu et al 2012), and by Liliana Rojas-Suarez from CLAFF². Bubbles are a serious concern despite Latin America's relatively low levels of
mortgage credit as compared to international standards.

Numerous policies have been put into place to try mitigate bubbles' harmful effects, but less has been done to prevent their development. This is partially a consequence of researchers and policy makers alike lacking yet a comprehensive general theory of their cause (Lind 2008). Despite great interest from policy makers in improving bubble regulation (Sarno 2010), these phenomena are diffcult to identify (Johansen et al 2001), which makes implementing controls during bubble formation particularly challenging. Preventive policies would be preferred because the events that unfold after the burst of a bubble are painful and panic makes regulation difficult.

Experimental approach to bubbles

Experiments are a useful way to test policies aimed at preventing bubbles (Palan 2010). Bubbles easily arise in experimental markets (Smith et al (1988), Caginalp et al (2001)). Literature emphasizes the role of imperfect information on the creation of bubbles. However, many experiments, like ours, show that bubbles appear even under full disclosure of relevant information about the assets and a common valuation. Availability of information does not seem enough to prevent bubbles: an effective bubble informational policy needs to do more than revealing it. A candidate channel is to aect the way information is perceived.

Mental Accounting and Bubbles

We propose that mental accounting can have an effect on the creation of certain instances of bubbles: investors have separate mental accounts for speculative assets (and their gains and losses), and cash and other resources that are considered risk-less and more liquid. Purchase price, instead of the discounted present value of the dividend stream, is often taken as the reference point for valuation of speculative assets. If an investor has purchased an asset for $1, and it goes up to $2, a gain of 100% will be registered. A decrease to $1.5, in a mental account that uses purchase price as the reference, still shows a 50% gain. Further decreases in value can still be registered as profits using this reference point, which increases the tendency of investors to keep assets regardless of fundamental values as long as they have shown gains in recent periods. Also, the separation in mental accounts prevent these losses to be completely internalized as a loss in cash. Different reference points (purchase values) for different investors would encourage trade even when preferences (i.e. risk aversion), utility (value obtained from holding an asset), dividends and information structure are common for all investors.

Our experiment

In other experiments, dividends and asset gains are usually distributed in the form of non cash credits: credit in an investment account or fake game money in experiments. As in the casino example, we believe investors place these non cash credits in the asset mental account and are more willing to risk them than cash. We propose a cashing out policy (convert the asset holdings and dividends, and give them to participants in real cash every period) that would make investors pool all assets in a single mental account and reset the reference point to the stream of dividends instead of the original purchase price. In Quintero (2012), we show that such a policy is successful in preventing bubbles in experiments. The setup is a standard double auction mechanism with 15 periods, based mainly on Smith et al (1988), that can be summarized in the following steps:

1. Assets and game money endowments are distributed, and information about dividends distribution is shown.

2. Bids and offers are received. Highest bids and lowest offers are shown. When a seller decides to accept a bid (a buyer accept an offer) the transaction is registered. At the end of period, dividends are realized and distributed.

3. In treatment versions, participants have their dividends and asset holdings converted to real cash every period. Then, participants are asked if they wish to convert that money into assets again or keep the real cash.

Table 1 shows the parameter values used in the dierent designs.

Column 2 and 3 show the number of participants in each experiment, the game money and number of assets received by each as endowment respectively. The 4th column shows the dividend values for each design with probabilities (.2,.3,.3,.2) respectively. E(d) is the expected dividend per period.
 is the intrinsic dividend value per share, which gives the expected stream of dividends each period (the table shows the value for period 1). This is the fundamental value that rational expectations theory predicts price should converge to.

Figure 1 shows the equilibrium price deviation from the fundamental value in each period. With inexperienced subjects, a clear bubble appears. In period 9th the volumes transacted signicantly decrease, which typically forecasts a burst. In the treatment version, when the cash out policy is implemented, the bubble practically disappears.

Figure 1: Prices exhibit larger price deviations from the fundamental asset value in the control groups vs treatment. Large deviations from periods 6 to 13 form a bubble in the control market that is partially prevented with the treatment.

Figure 2 shows weaker results with experienced subjects (who are allowed to play 15 periods previously without any real payffos). In this case, a smaller bubble in the control experiment is caused by speculators that bought assets at prices near the fundamental value. Some participants managed to sell and placed among the highest profit makers. Our policy lower the bubble, but it is much smaller to begin with. Its impact does not seem so relevant when the participants are more experienced. It is likely that learning took place and expectations of a bubble were used to make prots as discussed in Caginalp (2001).

Figure 2: Prices exhibit lower price deviations in the specication with inexperienced subjects. The policy still reduces the smaller bubble.

The setup with experienced and inexperienced participants is probably the most realistic (greatest external validity) (Huang et al 2006) and with more theoretical relevance³ In this setup, we obtain the largest bubble (Figure 3). Apparently, a fraction of inexperienced participants is enough to generate the strong effects that we assign to mental accounting. The cash out policy has the strongest impact in this case. Given the realism of this setup, the results make the stronger case for a potential application of a similar kind of policy in real markets.

Figure 3: The largest price deviations occur in this more realistic specication with both experienced and inexperienced subjects. The bubble observed between periods 8 and 14 is partially prevented with the cash out policy.

Policy Implications

Our results support the hypothesis that the creation of bubbles is affected by mental accounting. Mental accounting seems to affect the investors' willingness to hold or demand assets whose prices are unrealistically higher than their fundamental value. Our experiments show the appearance of bubbles even under full information disclosure. Learning seems to play a role, as larger bubbles are formed when at least some inexperienced subjects are present. We affect the way this information is perceived by implementing a cash out policy, which we interpret as changing the investors reference point and evaluation of gains and losses. Thus this policy would reduce their willingness to keep a position that would imply probable future losses despite a recent bullish trend in the market.

How can we implement this policy in real markets? Our experiments support the implementation of policies that emphasize the reality of asset gains and losses (including dividends). One possibility would be to manage the way information is presented, especially to less experienced investors, or those that do not have strong incentives to engage in speculative behavior. For instance, we could have speculative investment accounts joined to checking or current accounts, and have software combine the amounts of both into one hypothetical
grand cash amount. Similar measures could be tested in housing markets, although less smoothly than in other asset markets that are controlled electronically and more centralized regulated. Another possible regulation could be to require some time between transactions, especially with assets whose values are rapidly increasing. This would allow for speculative gains and cash to be placed in the same mental account. In the future, we could also use behavioral economics and experimental evidence to analyze policies that affect the behavior of winners and losers during the bubble, during longer investment horizons, and in complete markets that allow for actions like short selling. 

The implementation of policies derived from behavioral economics is most likely not the only definitive answer, but taking into account its lessons is useful to design alternative policies that help regulate such dangerous phenomena as bubbles, an objective that is today very important for Latin American markets.
 

* Tepper School of Business, Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213. Phone: (412)-304-4389. leq@cmu.edu.