Take an economic event or a sudden shift in price or volatility and compare it to a number of similar situations in the past.
Run your trade strategy through all these situations in the past while manipulating several variables. These variables include when to enter the market, the first and second profit targets, stop loss and risk/reward ratio etc.
Side note – Overfitting: A problem that can occur with AI is overfitting. Overfitting represents a model that correlates too closely with a particular set of data and it contains more parameters than can be justified by the data itself. Machine learning or AI tends to try to maximize its results on a set of data and ignores the suitability for future data.
Tradestation: These only work with price-driven trade ideas (if five consecutive green candles, then it will trigger a backtest). You basically pattern your system to recognize certain shifts in price or volatility and it will create a model that tries to adapt to similar situations in the future.
Building your own backtester: You can start from scratch and build your own backtester by various coding languages like Python or C++. It is not easy to build your own backtester because of various biases and how difficult it is to build an adequate model. Here is a website that explains more on how to build back testers.
Bettertrader: This is a tool that is built for backtesting news events and price-driven movements like for example oil moving up 3%. The advantage of using bettertrader is that a pre-existing model is built already for you and it only suggests statistically accurate trade-ideas to avoid bias.
In conclusion, backtesting is crucial for traders to gauge their strategy against historical situations before they engage in real trading with their strategy. One should be aware of when a model overfits its dataset by concentrating too heavily on a simple dataset. It’s necessary to be aware of how a model based on historical data might not predict a similar situation in real trading. While utilizing this tool, one should also combine several models to best represent the function that possibly represents future events in the market.