Hi all, I developed a system that I think might be ready for live testing. On paper, the stats look good. The test was done on a basket of 2334 stocks, from Jan 2016 to Dec 2021. I trade chinese equities, I especially picked that period because it had quite a few brutal moments. The system is long only (in China retail traders can't short stock). This is the equity curve These are the trades placed by the system total number of trades: 245 avg winner to avg loser: 1.70 win rate: 78.37% My backtesting system is very crude so I don't have access to many metrics, but I included the raw data of the backtest if anyone wants to check it out. The system backtests each stock individually so when putting everything together the results can be misleading, It's not as simple as summing everything up. my main concerns so far are that 1)there is a low number of trades per year, 2)all trades are clustered in very tight areas, and 3) there are moments when it goes completely flat. in the past, I had horrible result trading systems with perfect backtests, so I would like to have an opinion from someone else, what are the stats that I should look into? What are the most obvious flaws?
my main concerns so far are that 1)there is a low number of trades per year, 2)all trades are clustered in very tight areas, and 3) there are moments when it goes completely flat. ______________________________________________ Those should not be the concerns. It should be the fact of investing. Even if you trade, not all hours/days/weeks/months ....... are tradable.
5 years is a very long time. If you run a system long enough it will always yield profits. One test that I normally do is to break it down into smaller time periods, so you can see what would have happened every year, every 6 months, every month, every week and so on... Then you can see if you would have trusted a system that is not profitable for 6 months. When you say "I had horrible results with perfect backtests", to me that flags an over fitting problem. Have you tried to randomise your data to see if the system would perform that well?
Did you also backtest the actual trading interval to insure the backtest of the actual trading interval were identical (or believable)? If they do not match, your backtest is flawed.
update: tested also the last period (2021~2022 until now), results and equity curve are similar to the 2016~2021 period