My master's degree taught me MATLAB.
Not Python.
Since I didn't want to pay the $2,000 license fee, I watched every YouTube video I could find to teach myself.
96% of them were a complete waste of time.
The other 4% buried me in strategy ideas. Statistical arbitrage. Trend and momentum. Pairs trading...
Maybe that's you.
A head full of trading ideas but nothing really working.
You don't have an idea problem.
You have a sorting and filtering problem. Nobody showed you how to tell a real idea from a pretty chart.
Most published strategy ideas donβt work in real life:
- In 2020, Hou, Xue and Zhang re-tested 452 published anomalies. 65% failed a basic t-value hurdle of 1.96.
- McLean and Pontiff found that returns on 97 published predictors fell 58% after publication.
- Linnainmaa and Roberts tested anomalies on data spanning the 20th century and found most accounting-based return anomalies are spurious.
If peer-reviewed research from PhDs and professional traders fails this often, a 20-minute YouTube video will too.
Bottom line: every idea you backtest without sorting at best costs you a weekend and at worst, your trading capital.
As I always say: "algorithmic trading is a game of iteration."
Here's how to speed it up.
How to sort and filter your trading ideas: The 3-bet sort
Assets keep going, snap back, or move together. Sometimes we lose sight of this simple truth. Instead of fighting this reality, let's take advantage of it.
Step 1: Name the bet
Every strategy in those videos makes one of three bets. Momentum bets that winners keep winning. Mean reversion bets that a stretched price snaps back. Pairs bets that two related prices drift back together.
Frame the trade into one of these categories. Are you expecting an asset to keep going up (or down)? Then it's momentum.
Are you expecting an asset to reprice to some historic levels? Then it's mean reversion.
Are you betting on the relationship between two or more assets? Then it's a pairs trade.
Framing your trade can really help focus on what matters and avoid a lot of the noise that comes with trading.
Step 2: Run the one-number test
I call this the one-number test. Before any backtest, check whether the data leans the way you think.
This will help prove or disprove your hypothesis.
For SPY, cut daily prices into weekly returns and correlate each week with the next using autocorr(). Near zero means your bet needs a better reason than a chart.
Here's an example in Python of how to do this:
For pairs, use the spread np.log(a) - np.log(b) and swap pct_change(5) for diff(5).
Step 3: Backtest only the survivors
If the number agrees with your bet, write one rule and backtest it against buy and hold.
A great library to do this with is VectorBT. And with agentic AI, it only takes a few minutes.
For SPY that could be βbuy after a down week, sell a week later.β
If it disagrees, cross the idea off and work on the next one. Don't be surprised if most of your ideas don't pass this test.
Challenge: Change both 5s in the snippet to 21s. Does SPYβs bet change on monthly returns? If you can answer in five minutes, you just did strategy research.