TT Backtesting

Backtesting Considerations

Backtesting Considerations

About historical market data

The historical market data provided to the ADL algo is actual book data consisting of the bids / offers as well as the trades. When a backtest begins, the historical market data is replayed in the same time sequence as it occurred on the date that it was recorded.

When choosing historical market data, note that TT has data for most contracts going back to approximately late 2018. TT also continues to record market data going forward. TT will validate that the market data that you request is available before launching your backtest. In the first version of TT Backtesting, you will only be able to choose a maximum of one day’s worth of data to be replayed. This restriction will be relaxed in subsequent versions.

About replay speeds

When an ADL algo subscribes for market data for a given instrument, Algo Server attempts to deliver all market data updates to it. If the market data updates occur slower than the time it takes for an ADL algo to process them, Algo Server will send all of them to the ADL algo. If market data updates occur faster than the time it takes for an ADL algo to process them, Algo Server will conflate them. In this scenario, an ADL algo will receive the latest snapshot of the instrument’s bids/asks along with a list of all trades that occurred since the last delivery of market data updates to the ADL algo.

When selecting the replay speed for historical market data, bear in mind that using higher values may result in higher conflation rates. This may impact the execution of your algo. In other words, your ADL algo may not behave the same in a backtest as it does in production if you set the replay speed too high resulting in substantially different conflation rates.

For example, if your ADL algo relies strictly on trades that occur in the market or trades instruments which are not highly liquid, you may be able to set the replay speed higher without impacting the execution accuracy. If, however, your ADL algo is highly sensitive to changes in the inside market of highly liquid instruments, you may need to set the replay speed to 1.0. TT Backtesting also offers the ability to replay market data at speeds less than 1.0.

Users should be aware that replaying market data at speeds other than 1.0 may require changes to their ADL algos. For example, ADL algos that contain StopWatch blocks may need to be adjusted if they make specific assumptions regarding the speed at which market data is delivered.