One of the systems that has always appealed to me because of its simplicity is the breakout. In the breakout you’re buying strength and selling weakness so it makes intuitive sense. But how does it fare in real life? In this series of articles I’ll be looking at this simple system and investigate how one may optimize it.
Backtest and system rules
The backtest was done for the period 2009-01-30 to 2012-01-30 on 48 individual German stocks from the DAX and MDAX indices. The DAX in this period had both up and down trending as well as some sideways periods within a long-term up trend. The system is a simple 20-day breakout with no directional bias, meaning it goes short as readily as it goes long. Initial stop was placed at two times the ATR(10) and trailed with a stop just beyond the high/low of the last 3 days. Three exit possibilities exist:
- If the close of the day (first 3 days only) is lower (higher for shorts) than my entry price I would exit at the close.
- If a target of 2.5 times the ATR(10) is reached. This corresponds to roughly 1.2R (R = gain / initial risk, see Van Tharp Trade Your Way to Financial Freedom.
- If stopped out by the trailing stop.
For this article I’m only looking at the influence of the volatility. As described by Richard Weissman in his book Trade Like a Casino: Find Your Edge, Manage Risk, and Win Like the House, the ADX is a volatility measure even if mostly employed as a trend-strength indicator. In addition, I’ve examined my own volatility indicator, which is basically an instrument-independent normalized volatility measure that I call “A10”. Average value of A10 over time for any instrument is 10, lower values indicate low volatility, higher values indicate high volatility.
Below is a plot that summarizes the backtest and presents the A10 indicator. All parameters except Hold period and R-multiple are measured at the time of entry. Green represents the winning trades, red the losing trades. Starting equity was 100,000 EUR, results include 11.80 EUR round turn commission, but no slippage. A few things are evident right away: First, the majority of the trades are losers, which is to be expected for a system this simple with no setups, only one time frame, and one simple entry rule. However, this results in 4089 trades which is a good statistical sample! Secondly, the losses are mostly very small because of the aggressive exit rule (1), and the wins are all clustered around the target value 1.2R. One surprising fact is that this quite rough and simple system actually gives a slight positive expectancy of 0.02R.
The lower two plots show A10 for the long and short trades separately. One of the ideas suggested in Trade Like a Casino is that breakouts from low volatility situations are preferable because you can set a tighter stop, thereby getting bigger wins in terms of R, which is what really matters. This aspect cannot be judged from the below plot because the wins are capped at 1.2R. I do make another peculiar observation:
The A10-volatility distribution of wins is different between long and short trades, even though the losers have the same distribution! It seems like for the long trades the W/L-ratio improves at A10 < 8 and becomes close to 50/50 at A10 < 6-7, although the numbers start to get small. On the contrary, the short trades have a much broader distribution and the top appears to be shifted both with respect to the short losers and to the long winners. The cause for this is not clear to me, but could indicate that winning short breakouts in stocks come from a medium-high volatile environment, while the winning long breakouts come after periods of relative calm. Perhaps not a new or striking conclusion, but these data could be illustrating just that. The plot below shows the same kind of distribution only using the ADX(10), top plot using counts, bottom plots normalized. Here the distributions are much more similar and I would be tempted to conclude that either ADX(10) is not a good representation of the volatility, or at least that it is not a good way to differentiate between different market conditions. Other periods for the ADX might of course give different results.
Addendum: Richard Weissman correctly observes that this period is mainly a bull market. The next article will examine a period around the bear market of 2008.