ABG Analytics logo
 

Predictive Value of Two Technical Scores

Table 1 describes the predictive value of the traditional score over the 100 stocks in the Standard and Poor's 100 Index as of March 2002, using the prior 1,000 market days as of April 30, 2002. Given the technical indicators for each stock in each day in the sample, the table reports the average percentage return over the subsequent 20 market days as well as the standard deviation of the percentage return.

Table 1. Percent return over subsequent 20 days and standard deviation of returns, by traditional technical score

-------------------------------------
Traditional     Mean        Std.
Score           Return      Deviation
-------------------------------------
    1           0.016           0.150
    2           0.017           0.137
    3           0.004           0.116
    4           0.004           0.108
    5           0.004           0.107
-------------------------------------
In Table 1, we find that the traditional technical score has little value in predicting the direction of stock returns, at least over the period and stocks studied here. The average percent 20 day return ranged from 1.7% for those with a score of 2 to 0.4% for those with a score of 3, 4, and 5. Those with the lowest scores actually outperformed those with the highest scores, contrary to what the traditional interpretation of the underlying stock indicators would suggest! These figures suggest little or no value in short-term trading (with a 20 day horizon) after transaction costs are taken into account. In terms of evaluating short-term price direction (average returns), there is little value in this technical score.

In terms of evaluating risk as captured by volatility in returns, however, there is apparent value in this scoring system. Note that the standard deviation of returns is 15% for stocks with a score of 1 and about 11% for stocks with a score of 5. That is a meaningful difference in risk. According to these results, the real value in traditional technical analysis may lie in reducing risk rather than predicting direction.

Contrast these results with those of our statistically determined score in Table 2. Here, there is much more variation in average 20 day returns by score, and the average returns increase steadily as the score increases. Stocks with a score of 5 on average, gain 3.1 percentage over the next 20 days, and those with a score of 1 lose about -.3 percent. While 3.1 percent in 20 days is a nice average return, it is associated with high volatility and therefore high risk. The standard deviation of returns is 15.4 percent for stocks with a score of 5.

Table 2. Percent return over subsequent 20 days and standard deviation of returns, by statistical score 

-------------------------------------
Statistical     Mean        Std.
Score           Return      Deviation
-------------------------------------
    1          -0.003           0.115
    2           0.000           0.105
    3           0.003           0.111
    4           0.009           0.117
    5           0.031           0.154
-------------------------------------
This statistical score is much more reflective of the tenet of financial economics that a higher risk is required to capture a higher expected return. It also shows the difficulty of short-term trading. The lower-risk scores present no gain on average once reasonable transactions costs are taken into account. Buying stocks with a score of 5 may be the most profitable on average and over time, but at substantial risk. Of course, these scores may be combined with other criteria, with the aim of substantially improve the risk to reward profile. We use these other criteria in our trading systems.

A reasonable criticism of this analysis is that the score has been created to statistically fit data from the past. We should expect this scoring system to work as it did in Table 2 going forward as a best-case scenario, to the extent that historical relationships continue to hold. As relationships shift, as they likely will, this scoring system may not work as well. Note, however, when the sample was divided into two equally-sized periods, and scores were computed separately for the two periods, the two scores were highly positively correlated, suggesting some stability in relationships over the two periods. Furthermore, the scoring weights never get too out of date, since each week the weights are re-estimated with the addition of the latest week to the analysis data, and the dropping of the earliest week. The fact that a long period and many stocks were used bolsters the robustness of the scores.

How do the traditional technical score and the statistical score compare with each other? Overall, they are negatively correlated. That is, higher traditional scores tend to imply lower statistical scores. That is the fundamental principle underlying contrarian trading/investing strategies. For short-term trading, it is usually more profitable to buy stocks that have done poorly in the past than to buy stocks that have been increasing steadily in value. It also indicates the additional higher risk involved in applying some contrarian (countertrend) strategies. We disagree with some authors who suggest that adding stop losses to countertrend strategies necessarily defeat the strategy. While they do reduce expected return, it is a way to limit the inherent risk involved in such strategies.

Continue to part 5 - "Simple Trading System Performance - Moving Average Crossover Systems"

Contents: A View on Technical Indicators and Trading Systems