As one of my readers pointed out, it is now well accepted theory that stock prices exhibit a long-term reversal effect: stocks that experience long-term price depreciation tend to outperform the market over ensuing periods [and vice versa].

A commonly accepted explanation from behavioral finance is that because we are emotional beings we have a tendency to immediately throw Bayes’ Theorem out the window the moment we encounter dramatic and unexpected news. This phenomenon is not limited to stock prices but has been documented in other areas as well.

The seminal work on this is from the now famous paper from De Bont and Thaler (1985), “Does the Stock Market Overreact?” The paper proposed the simple hypothesis: “If stock prices systematically overshoot, then their reversal should be predictable from past return data alone, with no use of any accounting data such as earnings.”

To test the hypothesis, they took U.S. stock price data from 1930 to 1975 and created two portfolios (Loser Portfolio and Winner Portfolio) in each of 16 non-overlapping three-year periods. For the three-year period study, 35 positions were held. In the Loser Portfolio, the 35 worst performing stocks in the prior 36 months; in the Winner Portfolio, the 35 best performing stocks. For each three-year period, the cumulative average residual return was calculated for both portfolios. They then stacked the 16 non-overlapping three-year periods one on top of the other and calculated the average cumulative average residual return for the Loser Portfolio and the Winner Portfolio.

The results were quite impressive. Over a nearly 50 year study, the Loser Portfolio of 35 stocks, held for 36 months, outperformed the market by, on average, 19.6%. The Winner Portfolio was down about -5%, on average. The difference between the two portfolios was, on average, 24.6%, for 16 non-overlapping three-year periods. (t-stat: 2.20)

The other major conclusions from the study were: (1) for a formation as short as one year, no reversal is observed, (2) a large proportion of the future outperformance of past long-term losers over past long-term winners is found in January, (3) the winner portfolio has a higher CAPM Beta (1.369) than the loser portfolio (1.026), (4) the overreaction effect is asymmetric (i.e., more pronounced in the Loser Portfolio than the Winner Portfolio), and (5) they are surprised to conclude that Benjamin Graham was right: “the overreaction phenomenon mostly occurs during the second and third year…”

Of course I’m curious to know if this theory holds up against Japanese Net-Nets and to explore the question: Do Japanese Net-Nets with the worst near-term performance outperform?

To start, I pulled price return data for the prior 3 years as of portfolio formation. We could review this for 1-year returns, 5-year returns, or any other chosen parameter, but I decided to start with 3-year, as this was the focus of the paper. I chose not to use total return date (including dividends), but we could test for that too. (It didn’t appear to make much of a difference, but that was from a cursory glance.)

Instead of picking a set bucket size, I chose to break up the buckets, based on prior 3-year stock price performance, by tertile. T1_Worst is the bucket of stocks with the worst 3-year price performance, T3_Best is the bucket of stocks with the best 3-year price performance leading up to portfolio formation, etc. Below is a chart showing size (number of positions) by year.

Based on the three buckets created [Worst, Mid, Best], below is the average 3-year prior returns (price only) for each bucket for each year. The worst bucket (T1_Worst) had negative 3-year (cumulative) returns for almost every year studied, while the other two buckets (T2, T3_Best) had mixed results.

The first thing I wanted to check was the performance of a simple 1-year equal-weighted strategy. The portfolio is rebalanced every year in each of the 3 buckets. The performance shown in the chart below is unflattering as both the Worst and Best buckets have below average returns (compared to all Net-Nets), which is not to be expected. However, the meat of the Worst bucket’s outperformance shows up in years 2 and 3, as the paper theorized. So there’s not much to conclude for now.

To evaluate 3-year returns following portfolio formation is a bit tricky because this study is dealing with overlapping periods. But for a simplified approach, assume one contributed new capital to the portfolio each year (or at least for the first three years of the study). The first bucket in 2010 would be constructed and held for 3 years, then sold and rebalanced. The second bucket in 2011 would be constructed with additional capital and held for 3 years. And so on.

Below is a chart showing the 3-year mean (or equal-weighted) returns (including dividends) for each bucket for each year of formation. Again the returns are overlapping, so feel free to draw your own conclusions. But from this simple approach, aside for the first 3 cohorts (2010, 2011, and 2012) the T`1_Worst bucket does not show consistent outperformance when evaluated on a 3-year time horizon.

To review the results from another angle, below is the 3-year rolling mean for each strategy. Again the T1_Worst portfolio is strong out of the gate, but then considerably lags the average returns of Japanese Net-Nets over the length of the study.

Once you have a basket of Japanese Net-Nets the results of this very simple study suggest that it may not be a good idea to further screen based on the worst past price performance. Further study is of course warranted but there is little so far to get excited about.

Thanks for reading!