What Should You Consider While Backtesting A Strategy - Part 1
Backtesting is a very useful way to figure out how our trading algorithms might work in real life (might is the key word). But it can be hard for a data scientist who doesn't have a background in finance to understand what it all means. Sharpies or Sortinos? Returns or profitability with money? This shouldn't stop you, though, because some of the best funds in the world are run by people who aren't in finance. Instead, it's time to learn. In this article, we'll talk about some of the most important ways to tell if your trading strategy is working or not. If you really understand these basic indicators, you'll have a good basis for judging different strategies. Setting Up Performance Measures These are some of the first criteria or measures you could use to figure out how well your trades are going. Most of the time, the measurements focus on two important parts of a strategy: the change in the value of the portfolio and the risk of making those gains or losses. By understanding these two things, you can figure out what it does well and where it falls short. Financial metrics All of the metrics in this section tell you how much money you made (or lost) when you used a certain strategy. The final amount of money is a good place to start, but there are other signs that give us more information: Annualized Return: The average annual percent profit from your trading strategy (or loss). Win/Loss, Average Win/Loss: Total (or Average) (or Average) Profits from Trades That Work The total (or average) amount of money lost on trades that go wrong. % Profitability is the number of profitable trades out of all of them. When we talk about return on capital as a percentage, we usually mean that the strategy is a multiplier on your initial capital. This is helpful most of the time, but we should remember that it's only partly true. Next, if we want to fully understand a strategy, we need to know how we are making money. For instance, do we consistently make tiny wins or do we consistently make small wins followed by massive losses? By looking at different combinations of profitability, win/loss, and profit/loss, we might start to understand how our plan will work. Metrics that focus on risk It's just as important to see big profits as it is to know that the method could lose money in the long run. "No risk, no reward" is a saying that only winners use. The vast majority of people whose risky bets didn't pay off don't say it. Here, the following crucial metrics are important: Annualized Volatility: The standard deviation of the model's daily return over a year. Since volatility is used to quantify risk, a model with a higher vol indicates greater risk. Highest Drawdown: The most negative change in the value of the whole portfolio or the biggest drop in PnL. It is based on the biggest difference between the high and the next low before a new high is set. Since our backtest will always cover the whole period, drawdown is an important risk factor to think about, but we're much less likely to keep a losing trade open in real life. If you had bought Amazon stock in 1998, it would have been smart to keep every share and buy as much as you could during the dotcom bust. In reality, not many people would keep going with a deal if their money dropped by 10%, 20%, 40%, 80%, etc.