My notes on the previous chapters of the book here, here and here.
The ninth chapter discusses about the building blocks of trading such as breakouts, moving averages, volatility channels, time based exits and simple look backs in detail. The next chapter follows with a detailed discussion on various systems such as the ATR channel breakout, Bollinger breakout and Donchian trend etc. This chapter also gives the performance data for all these systems based on the historical data. For ex: donchian trend has a 10 year return of 30% p.a with a max drawdown of 38.7%.
The important point in this chapter is the author’s emphasis on backtesting. Backtesting means that every system should be evaluated with respect historical data for returns and maximum drawdown. Backtesting may not help predict the future or ensure that the system will always work, but it would help to determine which system could be profitable in the future and what conditions are needed for the success of the system.
My comment: The same approach should be applied by investors too. For ex: value investing has almost a 50 year history of performance over varying periods and business conditions. So this approach to investing has proven its ‘fitness’ over a long period of time and in varying conditions. I would say that any other approach such as momentum investing should also be evaluated in a similar manner.
The next chapter discusses in detail the pitfalls of backtesting. The key reasons why the historical test results differ from actual trading are as follows
- trader effects : As more traders use the system, the effectiveness of the system is lost
- Random effects
- Overoptimization paradox:
- Curve fitting: Fitting the system to data
The chapter then discusses how these distortions can be resolved and backtesting results improved.
The next chapter discusses how one can get better results from backtesting. One approach is by using better measures such as RAR (regressed annual return), R-cubed and a robust sharpe ratio. In addition a representative sample and appropriate sample size can help to get better results. The author also discusses about monte-carlo simulations to analyse the various systems based on historical data.