2-23. Intro to Backtesting

In this lesson, you learned about how backtesting fits into the quant research process, common mistakes that lead to misleading backtest, and best practices needed to produce a successful backtest. Now, that we’ve understood these important principles, we’re ready to start setting up a real backtest. Let’s get to it.

Great to see you again Gordon, I have a few more questions for you about AI and finance. How is AI being used in finance today? What applications have you seen used successfully or unsuccessfully? AI is being used in a lot of places. So first of all, some of the same technologies that are … Read more

Boosting, as we saw before, is a bit more elaborate. There are a few ways to do it but one of the most popular is this algorithm ADABOOST discovered by Freund and Schapire in 1996. Here’s the gist of it. Although we’ll develop the math in a bit more detail later. I must say, if … Read more

Proper statistical methodology in general and proper use of a test set in particular give a clear path detect overfitting. Essentially, when the model’s performance on the test set is noticeably lower than on the training and validation sets, then this is evidence that the model may have been over fit. Of course, it is … Read more

Overtrading is defined as trading in larger sizes than would be optimal. If the expected size of a mispricing or an arbitrage is 10 basis points, then no model should trade in such a way as to have 15 basis points of market impact, as this cannot be sustainably profitable. In extreme cases, overtrading can … Read more

Another way a back-test could benefit from hindsight and hence be an illusion, is if the modelers have statistically over fit the training data, and or failed to observe good statistical practices concerning testing sets. Typically, when statistical models are used, there’s a training set, a validation set, and a test set. Or in certain … Read more

A backtest is said to be valid or admissible if it satisfies at least the following two conditions. The first condition is that the profit calculation is realistic, which means that the simulated profit and loss could actually have been achieved by trading the same instruments in the same way that the backtest code implies. … Read more

A backtest is a simulation of running a trading strategy through some historical period of time. Ideally, the backtest should simulate as accurately as possible what would have actually happened in real-time production processes. If aspects of real-time production trading are left out of the simulation, then even after backtesting, there’s no reference for how … Read more

Welcome back. In the next several lessons,you’re going to learn about back-testing, which is how quantum simulate as rigorously as possible. What would have happened if a trading strategy was deployed in production over a historical period, in order to try to understand how it might perform in the future? As we mentioned earlier in … Read more