Thread by @theBuoyantMan

A small thread on backtesting parameters based on the tests I have done so far in this year (since march - on 10+ years of data)

1. Always test for 1 lot with no compounding. This will get you the actual max drawdown, returns, risk adjusted returns, and other required numbers.

2. Highly recommended to test for at least 10+ years. This will generate a good enough set of data from which we can derive the system parameters which we will base our decisions upon.

3. The first important parameter is maximum drawdown (otherwise called maxDD).

This maxDD number denotes how much maximum % loss your system made from its highs - measures the peak to trough % fall for every such fall, and gives you the highest such fall.

Why is this important?

It tells you how much you'd have lost in % terms.

The absolute number is important too, but the % terms number is more important. As the scale increases, the % number will determine just how much stomach churn you can take.

4. The next most important parameter in a system backtest is how long does it take for the system to get back to the prior equity high from its max drawdown point. This is important as you don't want to spend a lot of time (a year or more) underwater in terms of profits/equity.

5. Then comes the CAGR number. You will get the total return that the system has given for the number of years you have tested for. Most people just calculate the total number of points divided by no of years as a "no of points per year" kind of a calculation. This is useless.

What you should calculate is the compounded annual growth rate, CAGR. This shows you how your profits will be used to compound your account, and this number will also play a significant role in calculating the CAR ratio which will help with decisions regarding system comparison.

6. Win Rate - total number of winning trades / total number trades is the win rate. In a backtest of 100 trades, if your system won 40 trades and lost 60 trades, your win rate is 40%.

Note: A super high win rate isn't by default great as a standalone measure, and vice versa.

7. Average win / loss - this is an absolute number in currency figure. This will help you calculate the Risk vs Reward ratio. Most people calculate this as average win vs average loss.
A better way to approach this:
(Avg win * no of trades won)/(average loss * no of trades lost)

8. Once you have the RR figure, you have the basic parameters required to evaluate a system for tradability.

There are many more parameters - sharpe, mar, skew, etc., that you could look into and fine tune your approach. But for beginners, you could start with the ones above.

9. Your system's max DD should be lower than your CAGR. This is not absolutely necessary, but it's better you filter out such systems.

10. Systems with CAR ratio (CAGR / maxDD) of more than or equal to 2 are better. Here CAGR and maxDD are both % term basis points.

11. Systems with win rate between 40-55% are better to trade initially. Too high win rates like 70% or above may have negative skew - make money most of the trades, and the one or two trades you lose, you lose big. There's a famous saying - eat like ants and shit like elephants.

Too low win rates like 20-30% (typical with hardcore trend-following strategies) require a lot of heart and a strong stomach. These come with significantly high RR, but good potential for drawdowns too. Not really suitable for a beginner unless your balls are made of steel.

So, target systems between 38-55% win rate to begin with - the sweet spot where beginners can find an edge worth sticking to. Coupled with a decent RR and CAR figure, you could be looking at a tradeable system.

The negative skew systems are usually directional option selling based - that make 60-70% win rate, but when they lose, they lose big. Not advisable for people starting with <50L capital and definitely not for beginners.

12. For example: A system with an RR of ~1 : 2.5, with a win rate of 45%, and a CAR figure of 2.2 or something like that is worth looking into. Unless you test multiple systems, you won't know which one is worth taking through to live execution.

For this reason, don't stop when you find a decent system. Try and find one or two more systems to compare with, and pick the best one that suits your psychology, approach, and time restrictions.

13. You can calculate all of these parameters after you have done manually backtesting 10 years worth of data or use excel/vba/python to test.
Mark the entry, exit, stop loss, p&l, commissions, etc., and then work on these calculations.

14. Be very generous with commissions + slippage that you allocate for the system and your system has to turn up numbers AFTER including slippage + commissions into the system execution, not BEFORE. Be wary of that.

15. Most systems that have glorious backtests fail in live trading because of issues related to execution - these could be timely execution, spread in that particular instrument, slippage, wrong choice of strike, and so on. Account for all these factors within your backtest.

16. Essentially you must treat your backtest like you'd treat your live trading process. If you're automating, treat your backtest exactly as your automated system would trade in realtime. If you manually trade, treat backtests the same way you trade and execute. No differences.

Where differences creep in between backtests and system execution is where deviation from backtests creep in, in the real-time trading results.

Hope this helps you to get started.