Over the past weeks I have been weighing up trading strategies.
I have identified two major issues which people always overlook.
Firstly, performance consistency. Averaged results over long periods of time often paper over long periods of poor performance.
Researchers report max drawdown, but I feel reporting the large losses in the context of larger overall gains is dumb (albeit an winning sales tool!).
What is more useful is an understanding of how long you may have to wait for the strategy to turn a profit.
Time is money!
I am introducing a 'Max Wait' analytic to the Lazy Backtest IDE, which reports the maximum number of years needed to ensure a profit.
For example, if I have 50 years of backtesting history. Perhaps out of 12 four year periods, one has incurred a loss. But out of 10 five year periods none have incurred losses. We would report the 'Max Wait' to be 5 years.
I.e. based on historical returns for this strategy you may have to wait up to 5 years to see a profit!
Note over 50 years, the 'Expected Wait' for the S&P 500 is 1 year (over 50% of 1 year returns have been positive) and the max wait has been 8 years (there have been 7 year periods which accrued losses!).
Based on the t-distribution, I have also included a 'Probability of Yearly Loss' analytic.
The other issue is comparability of performance results. Longer backtesting periods usually produce much lower yet sturdier Sharpes.
Standardising Sharpes to a 10 year lookback period gives a good idea how strategies with different backtest periods compare, removing any ambiguity!
These three new analytics are a decent step forward.
In fact I am wondering whether I should remove the non-standard Sharpe ratio all-together...