Something that should be obvious at this point is that the quality and quantity of the historical market data available is a key aspect in being able to build a profitable algorithmic trading business. For this reason, most market participants invest a lot of resources in building a market data capture and normalization process that is extremely accurate, and software implementation that is bug free and able to faithfully capture and replay live market data in historical mode to match exactly what algorithmic trading strategies will observe when they are deployed in live markets. Usually, if trading strategies are not performing in live markets as expected, this is the first place to start. By adding an extensive amount of instrumentation/recording to what market data update trading strategies observe, and comparing what is observed in simulations and live trading, it is relatively straightforward to find and fix underlying issues.

There may be issues in the historical market data recording setup, the live market data decoding and delivery setup, or both. Sometimes, latency sensitive trading strategies have a normalized market data format in live trading that is different from what is available in historical recording by streamlining market data information delivered to live trading strategies to be as compact and as fast as possible, in which case this can be another reason why live market data updates differ from historical market data updates. If issues are discovered in this step, first fix those issues in the historical and/or live market data protocol. Following that, the trading strategy results are recomputed, recalibrated if needed, and then redeployed to live markets to observe whether fixing these issues helps to reduce simulation dislocations.