Double Then Nothing: Why Individual Stock Investments Disappoint

351774_more_dice_series_1This paper tests the argument by behavioral researchers Tversky and Kahneman (1974) in their seminal work, that individuals often rely on heuristics or rules of thumb that reduce the complexity involved in predicting values, but such heuristics can lead to severe and systematic errors. I test their argument in the context of investments by focusing on a simple heuristic whereby investors are attracted to buying stocks that have recently doubled in price in anticipation of further gains. Such a study is important because relying on such simple heuristics may destroy investors’ wealth. I show that a strategy of buying stocks that have recently doubled in price can lead to predictable disappointment for these investors and severe underperformance relative to the market (-28% over a four- year period), whereas investors who avoid relying on this simple heuristic are likely to perform as expected, on average similar to the overall market. This ―doubling‖ variable is a significant predictor of future price reversals in addition to past performance per se, as uncovered by DeBondt and Thaler (1985) in their well-known overreaction study. Thus based on this study investors can become aware of the dangers of relying on simple heuristics and can avoid disappointment in investment returns.

This paper’s research focus on stocks that have doubled in price is motivated by axioms highlighted in previous research, investment books, and the popular press due to its simplicity, since any investor can readily relate to and strive for such doubling performance. For example, Reinganum (1988) identifies ―winner stocks as those that have doubled within a calendar year, with his sample firms drawn primarily from William O’Neil’s publication, The Greatest Stock Market Winners: 1970-1983. Previous studies suggest that investors may be influenced by perceived price trends. DeBondt (1993) argues that, besides fundamental explanations, there are two other possible explanations as to why stock prices fluctuate, both of which are related to individual investor psychology and systematic misperceptions of value. First, investors put too much emphasis on the latest information and not enough on base-rate information, an application of Tversky and Kahneman’s (1974) representativeness heuristic. Second, investors tend to discover trends in past prices and expect such trends to continue.

DeBondt (1993) experiments by giving subjects 48 months of past prices for a variety of series and asks them to predict prices 7 and 13 months in the future. Based on 38,000 forecasts of stock prices and exchange rates he finds that non-expert individual investors expect a continuation of apparent past trends in prices. More recently, He and Shen (2010) estimate expected returns directly from stock prices and financial information and show that investor expectations are overoptimistic for stocks that recently experienced high returns.

This study attempts to replicate the data-gathering behavior and performance of some positive feedback3 or momentum traders who follow a simple price-trend heuristic to make investment decisions. I begin with a sample universe that contains a high proportion of stocks with a recent ―stellar past performance. To avoid any ―penny stocks a minimum stock price of $5.00 is required at the beginning of the screening period. The primary screen focuses on the stock’s recent track record. Despite the typical disclaimer that past performance is not indicative of future results, as the studies above suggest, past performance is frequently used (for example, by positive feedback traders) as at least one important investment criterion. I argue that of particular appeal are any stocks that have doubled in price in the recent past, which I arbitrarily define as within the last four years (DeBondt (1993) presents subjects with four years of historical data). Identifying that a stock has recently doubled in price is a simple reference point for an individual investor, much simpler, say, then identifying a stock as being in the lowest decile of returns within a particular dataset over a particular sample period (as is common in many studies) – in the former case, all that is required is the recent price history of that one stock while in the latter case one needs to make a relative comparison over a much larger sample. I screen on month-end stock prices for up to 48 months. If a stock has doubled in price within that time period, then it is categorized as a ―stellar stock that should appeal to positive feedback traders and is immediately placed in the investment universe (e.g., if a stock doubles in price after 18 months then no more history is required). Momentum stocks would typically fall under the stellar stock category so long as the stock has doubled reasonably quickly (e.g., in a 12-month period).

The other performance-based category is ―non-stellar stocks, i.e., those that do not double in price but yet still have a complete four-year track record. Such stocks might form the universe for all other investors, whom I refer to as the fundamentalists. Value stocks would typically fall under this category. Note that any stock with a shorter track record (e.g., because it has gone bankrupt or has no longer met the listing requirements of the exchange) is not included in either investment universe and thus a ―backward- looking survivorship bias is induced in the screening period. However, as I discuss below, there is no survivorship bias in the testing period.

In this study, I find an almost even split of the stocks in this survivorship-biased screening period sample that have at least doubled in price versus those that have not, with a total sample (i.e., doubled and non-double stocks) median annual return of 20.1% or a median excess-of-market return of 10.1%. However, in a subsequent four-year (survivorship-bias-free) investment period, only about a quarter of the total sample stock prices doubled (or more), with a total sample median annual return of a disappointing 6.6% (excess-of-market return of -3.6%). Those that doubled in the screening period are less likely to double subsequently than those that had not doubled previously, invariably leading to disappointment for the positive feedback trader group. The cumulative excess return after four years for those stocks is -28.0%. In contrast, fundamentalists who invest in stocks that did not double during the screening period experience near-zero cumulative execs returns after four years (-0.2%).

I then investigate the extent to which stock returns for this sample are predictable and thus whereby investors can improve their chances of investment success. Much of the cross-sectional variation in investment period returns can be explained not only by past stock performance (a negative relationship) and test period (or investment period) market returns (a positive relationship as expected), but also whether the stock has recently doubled in price (negative), past earnings (a positive relationship), and various valuation- related metrics measured at the start of the investment period. A probit model identifies ex ante variables that are able to predict whether or not a stock will at least double in value over the investment period. An investment strategy based on the predicted probability of a stock doubling offers large potential rewards.

While this study is related in particular to the overreaction or contrarian profits literature and papers such as DeBondt and Thaler (1985, 1987), it is nonetheless distinct in a number of ways. First, instead of focusing on categorizing stocks in portfolios based on historical ―winner‖ or ―loser returns relative to one another, it relies on one simple heuristic readily available to any investor with a recent history of past stocks prices – identifying whether a stock has doubled in price within the past four years. Second, this study relies on a much more extensive sample of firm-observations, including over 5,000 cases of firms that have doubled during the screening period. In contrast, DeBondt and Thaler (1985) focus on portfolios that average only between 35 and 50 stocks for their three and five year periods, respectively. Third, as I show in a hypothetical example of price patterns in Figure 1, it is not necessarily the case that the firms that I categorize as ―stellar‖ (and thus have doubled in price over the screening period) coincide with DeBondt and Thaler ―winners. As I show in the figure, it is possible that my sample of ―doubling‖ stocks might actually include stocks that would have been categorized by DeBondt and Thaler and others as either winners, losers, or in neither such category. In this example, all three stocks have an end-of-screening period price of $20, and all have doubled in price in the previous 12 months, thus being categorized as the stellar ―doubling‖ stocks in my sample. However, over the entire 48-month period, three different patterns emerge with the ―loser‖ stock dropping from an original price of $40, the ―winner‖ stock increasing from an initial price of $5, and the ―neutral‖ stock fluctuating around $20. Thus while some of my results are consistent with some of the findings of previous studies, I argue that the phenomenon of the doubling stocks is an example of a simple heuristic and may be distinct from the winners/losers phenomenon in a similar way that Thomas and Hwang (2004) find a 52-week high phenomenon that is related to but distinct from other momentum studies. Read the full paper.