I want to suspend for now the series of posts regarding volatility (may go back to it some time in the future), and discuss an often overlooked but extremely important topic in developing a trading strategy that works outsample and is not just curve fitting: performance metrics.

While I consider Sharpe Ratio a fairly good metrics, I don’t think alone it’s enough to tell you how good (or bad) a strategy can be: an example is a strategy with an high Sharpe Ratio but with long periods of drawdowns, which may not be a feasible strategy to trade live (at least, not as a standalone). Statistical arbitrage strategies, for example, and in general all the strategies with highly not normal returns distributions can show this kind of behaviour, as also pointed out by Dr Chan and Jev in their respective blogs:

Quantitative Trading – What are we to do with Sharpe ratio?

Quantum blog – When Sharpe is useless

One way around this is to look for other features in a strategy besides high Sharpe Ratios, such as high percentage of winning trades, the average win being bigger than the average loss and small and short drawdowns. Related metrics are:

**Win %** = # of winners/ total # of trades = P(W)

**Average win
Profit Factor**=

*sum winners/ sum losers = P(W)* average win/(P(L) * average loss)*

**Expected Gain**= P(W)* Average win – P(L)* Average loss

**Average # of trades**in one time unit

**Max % Drawdown**

**Longest Drawdown**

Particularly useful and all-inclusive are the Profit Factor and the Expected Gain.

Once these metrics are calculated, one simple way to summarise the results is to build a fitness function which adds them together using certain weights, decided according to the user “utility function” . In doing this, particular care must be put in scaling the different figures in a sensible way, so to make them comparable.

Keeping track of these metrics is of help in discerning suitable strategies and can give you an insight of where your edge, if any, is (you can have, for instance, a strategy with a low winning percentage but a very high average win compared to the average loss…).

To conclude, I need to specify that all these figures have to be *estimated* somehow and since this is not a bet with well defined wagers, nothing guarantees us that these values will remain constant over time (and actually in general we can be pretty sure that they *will* change): at the end of the day **common sense**, **inquisitiveness **and **intellectual** **honesty **are the best weapons in a quant trader’s arsenal.

Andrea

Pingback: Perils and rewards of Diversification: improve your system’s Sharpe | Math Trading