EP04 - Alpha Decay

Why trading strategies decay over time. Learn about backtesting mistakes, overfitting, look-ahead bias, reverse engineering, and why alpha strategies become obsolete.

Tweet coming soon

I'm retiring! I'm buying an island!

My Python script turned $1,000 into $10 Million in the simulation!

It's a simple Moving Average Crossover. I downloaded a library, plugged in the data, and optimized the window size.

The perfect window is 14.2 minutes. It never loses!

`price = df['close']`.

`if price > average: buy_at(df['open'])`.

Genius strategy! You gotta pay me to license it!

You are using the `Close` price of the bar to decide whether to buy at the `Open` of the *same* bar.

So?

The `Close` happens *after* the `Open`!

You are deciding to bet on the horse *after* you watched it win, then traveling back in time to place the bet at the start.

That is **Look-Ahead Bias**.

Oh. So... I cheated?

You built a causality violation engine.

Fix it. Use `shift(1)`. Make decisions only on *yesterday's* data.

Okay, $1 Million profit. Still good!

The strategy is still solid!

You tested 5,000 different parameter combinations and picked the best one, right?

Yeah! Optimization!

Looks like you gotta lose some money to learn this lesson.

You didn't find a strategy. You tailored a suit that fits *this specific mannequin* perfectly.

You stitched it so tight it can't move. This is **Overfitting**.

The moment the market moves differently than the past (which it always does), your "optimized" strategy explodes.

So... how do I know if it fits a real human?

**Out-of-Sample Testing**.

Train on 2018-2020. Lock the parameters.

Test on 2021.

If the line goes up in 2021, you might have something.

It... it died.

It reverted to the mean. You mined noise, not signal.

So the library lied to me.

The library did exactly what you asked. It found the best curve to fit the past.

But the past is a ghost. You can't trade ghosts.

And even if it worked... look at the fees.

Death by a thousand cuts.

I'm not giving up!

In 2 days I will find a strategy that actually makes money! You'll gonna bet with me on that!

Heh. Bet. You really had turned into a degen gambler.

Hahahaha! I found a strategy that works! A real one!

I locked it in this cold storage vault. No one can steal the code. It's safe forever! I can set and forget and print money forever!

Safe? Shez, look at the floor. Your profit is leaking.

It's melting! But how? The door is locked!

It's not melting because of the heat. It's melting because of **Alpha Decay**.

In Quant Finance, you don't own the patent. You only own the **Lead Time**.

The moment you start trading, you are telling the world your secret.

But I'm using encryption! VPNs! No one sees my code!

They don't need your code. They see your **Footprints**.

Every time you buy, you move the price up. Every time you sell, you push it down.

The wolves don't know *why* you are walking there. They just know that if they follow your tracks, they find diamonds.

Sophisticated firms analyze the tape. They see your pattern. They replicate it.

Then they execute it *faster* than you.

So... by using the strategy, I destroy it?

The Market is an **Efficient Learning Machine**.

An "Arbitrage" is a mistake in the market. You are paid to fix it.

Once it's fixed... you stop getting paid.

This sucks. Fintech companies build a payment app once and charge fees for 50 years.

Why do we have to reinvent the wheel every Tuesday? I should just quit being a quant and work for fintech instead.

That is why we don't protect the *Product* (The Strategy).

We protect the **Factory** (The Research Pipeline).

We assume every strategy will die in 6 months.

So we build infrastructure to test, validate, and deploy *new* strategies faster than the old ones die.

It's a treadmill.

It's the **Red Queen's Race**. You have to run as fast as you can just to stay in the same place.

You want to protect your profit?

Don't hide. **Innovate**.

Find new data sources. Faster FPGAs. Better math. More creative market exploitation.

It sounds exhausting.

Passive income is a myth. Every dollar is earned by someone's hard work.

Okay. New strategy. I'm creating fake immigrant profiles to apply for government grants and income.

That's **regulatory arbitrage** and it means it's illegal.