Hello, i've been backtesting a strategy on 12 currency pairs. i've tested as far back as two years for most. It's a fairly simple strategy that generates an average of 34 signal days per pair, per year. Long short it doesn't matter.The accuracy varies widely however. On one pair it's 47% accurate on another it's 27%. But even the pair with 27% accuracy shows a profit, a small one, but nevertheless. Anyhow, my main concern is that the majority of profits come from a handful of trades, if it would not be for the handful of big winners the strategy would only generate commissions and be flat by years end. My question is, would you trade this? Is it a viable strategy? thanks for the feedback.
that to answer for this quenstion was advisable hear more perfomance of this systems. This it is important to watch a equity curve. But if honestly-for me it's in a complicated way to trade with SMA. From Russia with love
Cant go past that, really. I despise backtesting as a rule, but it might be helpful, to begin with; Which currency pairs? Trying to avoid huge USD correlation would be tricky i'm sure. And two years isn't very long; currencies trend, sure, but 32 trades a year (on average) may be more than is optimal. The worthwhile swings/moves will be longer than that, generally. Chances are, the stop-loss mechanism is out of whack. Not that i beleive in backtesting, period, but there it is, just some thoughts.
I use more than two years historical price data to use for my testing. You might consider how your strategy might perform under the worst possible conditions such as a long continuing losing streak or a sudden catastrophic event such as the destruction of the World Trade Center when the internet was jammed, useless for trading and brokers might abandon their desks. Whether or not you use the system depends on your own needs, not mine. Are you comfortable with the volatility the system shows? Are you comfortable with the overall growth of the portfolio? When trading a portfolio the performance of an individual security is not important, the behavior of the entire portfolio is what counts. Trend following systems typically show only a small number - maybe 1 % - big winning trades. Those big winners generate most of the overall profit. I find the Richard Dennis interview in Market Wizards first volume to be a good text for studying trend following practices. If you are trading currencies then you might want to consider the effect of interest rate daily debits and credits to your account. In some cases interest rate debits and credits may be greater than trading profits and losses.
Different pairs have different trendiness and volatility characteristics, so no wonder you get such results. Could it be you strategy is trend-following only? or counter-trend only? In this case profitability relies on a few outliers. It's a normal thing in trading, but question is: does you system has an edge? It seems it depends on non-normal events that don't fall into normal distribution, so if you're not losing money when no such events occur then you found a robust profitable method. Also, two years of backtest may not be enough. But as you say, it generates 2x12x34=816 trades, it should be statistically significant data set. But beware that those two years might not cover most of market conditions. It could be, for example that these two past years were profitable because of some market condition, but if you took four years you could get worse results because those other two years had different market conditions. Try to include as many different market conditions (trending/non-trending, volatile/non-volatile) into your test data sample as possible.
I always worry about "data fitting" when only a couple of trades account for most of the profit. If it isn't profitable if you remove the best 5% of the trades, I would be very careful.