Do patterns formed by Japanese candlesticks, which summarize asset price behavior with a candle and two shadows indicating open-high-low-close prices over a given interval, work as intraday technical trading signals? In their August 2012 paper entitled “The Intraday Performance of Market Timing Strategies and Trading Systems Based on Japanese Candlesticks”, Matthieu Duvinage, Paolo Mazza and Mikael Petitjean investigate the power of 83 Japanese candlestick rules to predict intraday returns of the 30 components of the Dow Jones Industrial Average (DJIA) based on both stock timing metrics and optimized trading systems. They explicitly correct for data snooping bias that derives from testing a large number of rules on the same data and account for trading frictions. Using 5-minute intraday high-low-open-close prices from April 1, 2010 through April, 13 2011 for the 30 DJIA stocks (20,550 observations per stock), they find that:
- Evaluating rule timing performance 50 minutes after pattern appearance (or longer if the pattern repeats), with price trend measured over the 50 minutes before pattern appearance for rules that depend on trend:
- The number of signals per rule ranges from zero to 22,649 over the entire sample period.
- Bullish signals perform better than bearish signals. Based on gross returns, 56% (22%) of trades based on individual bullish (bearish) patterns are profitable, and 64% (39%) generate positive average returns.
- Correcting conservatively for data snooping bias, 26 (27) of 83 patterns are still significantly profitable based on average gross return (gross Sharpe ratio).
- However, only five (three) of 83 patterns are significantly profitable based on estimates of average net return (net Sharpe ratio).
- Results are generally robust to varying performance measurement parameters.
- Using a different way of correcting for data snooping bias, no rule beats buy-and-hold across the 30 stocks on a net basis for one-way trading friction of 0.05%.
- Of 2,047 complex automated trading systems developed by identifying the top ten patterns for each of the 30 stocks, picking the 11 patterns that are significantly profitable for at least two of the 30 stocks and combining the 11 patterns in different combinations, none significantly outperforms buy-and-hold after accounting for trading friction.
In summary, evidence indicates that, although some Japanese candlestick patterns exhibit short-term predictive power, traders cannot reliably exploit these rules for intraday trading of large and liquid U.S. stocks after accounting for luck, risk and trading friction.
Cautions regarding findings include:
- Technical trading rules may be more effective on less visible stocks (more likely to experience inefficiency). However, trading frictions would also tend to be higher for such stocks.
- Combining candlestick patterns with other technical indicators may be more successful. However, such combinations impound incremental snooping bias.
