Does technical market analysis work? In their June 2014 paper entitled “Technical Market Indicators: An Overview”, Jiali Fang, Yafeng Qin and Ben Jacobsen examine the profitability of 93 market indicators as applied to the S&P 500. Of the 93, 50 are market sentiment indicators that attempt to predict market behavior based on the supposition that stock prices tend to rise (fall) when bullish (bearish) sentiment dominates. The remaining 43 are market strength indicators that attempt to predict market trend continuation based on breadth of movements as indicated by volume, number of advancing/declining issues and number of periodic highs/lows. 65 of the 93 indicators are raw (such as numbers of advancing and declining stocks per day), and 28 involve measures constructed to suppress noise (such as number of advancing issues minus number of declining stocks). The authors use the S&P 500 as a test market because of its long history. They consider entire sample periods, equal subperiods, different economic regimes (expansion or contraction) and different sentiment regimes (bullish or bearish as indication of degree of investor irrationality). They employ a generous 10% significance level for statistical tests, with and without estimated trading frictions of 0.10% for switching between the market and a risk-free asset. Using the longest samples available for each indicator through the end of 2010 or 2011 (averaging 54 years and as long as 200 years), they find that:
- Over the maximum available sample periods, 30 of 93 indicators exhibit gross predictive power. However:
- Only 10 of the 30 are robust across equal subsamples.
- Only eight of the 10 are stable based on rolling window regressions.
- After accounting for switching frictions, none outperform buying and holding the S&P 500 based on either Sharpe ratio or Jensen’s alpha.
- As determined retrospectively via NBER cycles, 26 (21) of 93 indicators exhibit gross predictive power during economic expansions (contractions). However, after accounting for switching frictions:
- During expansions, only one indicator beats buy-and-hold based on Sharpe ratio, and one other beats buy-and-hold based on Jensen’s alpha.
- During contractions, none of the indicators beat buy-and-hold.
- Using real-time Chicago Fed National Activity Index data rather than NBER data to define economic cycles, no indicator exhibits significant predictive power during expansions or contractions.
- Using the Baker-Wurgler model to define sentiment cycles, 21 (25) indicators exhibit gross predictive power when sentiment is high (low). However, after accounting for switching frictions, none are significantly useful.
- Findings are similar for the following additional robustness tests:
- Further noise reduction modeling.
- Controlling for or eliminating outliers.
- Excluding the 2008 financial crisis from samples.
In summary, evidence from comprehensive testing of technical market indicators on S&P 500 data offers little support for belief that they are useful to investors.
Cautions regarding findings include:
- The assumed level of switching frictions is very low for large parts of the sample periods used (see “Trading Frictions Over the Long Run”).
- Return calculations are for capital gains only (dividends are not included), giving an advantage to technical trading over buy-and-hold.
- The study does not account for the data snooping bias accrued from testing 93 indicators on the same set of returns. Such bias argues for stringent significance thresholds.
- Technical indicators may be more effective in markets less efficient than that represented by the S&P 500.
