From 2016 until 2021, I worked as a developer in quantitative finance with clients who took an interest in mining for profitable trading strategies using a backtesting engine with modifiable strategy parameters. For these clients, I built tools using Python and SQL to simulate a real trading environment, including order executions with broker specific trading costs and order types—trailing stop loss orders among them. Backtesting frameworks were complete with summary analytics. Reporting included strategy profitability, drawdown risk, and other measures the client personally found useful in ranking strategies. Some of the major tasks in implementing a backtesting framework included:
- developing a mechanism for storing, retrieving, and regularly appending new candlestick pricing data as it became available;
- accurately modeling market behaviors;
- developing analytics reports with useful ranking metrics;
- optimizing the speed and processing usage to run 5-10 year backtests in large batches to hone in on profitable strategy parameters quickly;
- and finally, introducing trading strategies into a live context, using identical code as was used in our backtesting framework to ensure that the results found in our analytics reports matched as closely as possible (barring data discrepancies, which are unavoidable it seems) in the live context as they would in backtest, preserving the integrity of our backtesting framework.