#84: How to Backtest a Strategy

S2N Spotlight

We are all on a learning journey. Something has been bothering me for some time and it is something I used to promote. Furthermore, this week I introduced it into this newsletter. I never felt comfortable about it, as it was a great example of the lethal power of statistics. Let me explain by way of a real example.

I really find Twitter (X) interesting but I simply cannot keep up with all the click bait that is sprouting on the platform. The tweet below is from a guy I have picked on before, and expect him to block me pretty soon if he hasn’t already.

He has presented a trade setup where the 5-day RSI of the Nasdaq 100 goes above 95 you buy the index and hold it for a month. You can see his results below.

He gets 10,400 views and I get 40 talk about feeling small. My analysis produces different results. I counted 32 signals over the period, to his 16. I produce a 4.01% one-month average return; he produces 2.9%. However, this is not the issue; I feel it is potentially misleading to provide statistics like I have done with the average returns 1, 3, and 6 months after a signal. Before I explain why, let me share a proper backtest.

Since I started Signal2Noise, I have been reporting backtested results following a signal. Anyone who has built a systematic trading system will have experience working with backtests. If a backtest shows wonderful results, you should always be wary. Let’s look at the above setup in a backtest.

Buying the Nasdaq and holding it for a month when the 5-day RSI is above 95 results in a 0.68% cumulative return and a pithy average return of 0.05%. You will notice that I count only 13 trades versus my count of 32 above. The reason is because I have a rule in my backtest that I can only trade one signal at a time. What typically happens is that when the RSI goes above 95, it will do it for a few days consecutively. It is not practical to trade every time a signal is generated, as you will run out of money or have way too much risk.

In case you are wondering, why don’t we simply do the reverse? In other words, sell the index for a month when it goes above 95. I have got you covered, but the results are even worse.

In summary, backtesting is far more realistic than providing a static snapshot of averages. When you trade time series, you have a compounding effect that is not taken into account with a snapshot average. For example, if you lose 50% in your first trade, you need to make 100% to be at 0%. If you make 100% in your first trade and lose 50% in your next trade, you will be up 50%. If you work with averages and you have 1 trade that loses 50% and 1 trade that makes 100%, your average return is 75%.

There is another important difference which doesn’t feature in the above example but is a factor in many examples. When you backtest, the trade goes on a journey where it can, for example, during the course of a month have a significant drawdown and still turn out positive. If you apply stop losses, then your backtest result might have a negative result for a particular signal, and the average will assume the snapshot at the end of the month and not factor in the reality of the journey.

S2N Observations

I remain bearish I was hoping to see that the breadth of the cumulative advance decline line of the Nasdaq was diverging. Call me early or just call me a romantic, but I see this setup as a kiss goodnight.

Timing the market consistently is impossible. However, I believe that positioning for a pullback very soon, if not tonight, is a good risk-reward strategy. I am anticipating a multi-month pullback with the potential for much more.

The relationship with the 2-year Treasury leading the Fed’s overnight interest rate is one I don’t like to bet against. If bond yields continue to drop, the Fed will have no choice but to cut interest rates. People are mistaken to think the Fed Chairman and the governors set the interest rate. The market is far more persuasive than the Fed chair. We need to be careful about what we wish for.

I try not to get carried away with major headlines. They are great click bait, and I could do with a few more clicks . Credit card debt in the US just crested over $1 trillion. This is more noise than signal; look at the chart below this one for more signal.

I look at the year-on-year growth of credit card debt. It is not growing nearly as much as it has in the past. I have already taken too much of your time. I think tomorrow I will take a look at how the stress on repayments on the credit card debt is looking, it is probably more insightful than a big headline round number of $1 trillion.

Performance Review

Chart Gallery

News Today

The post #84: How to Backtest a Strategy appeared first on Signal2Noise.