Research
You Skipped a Step
Your backtest worked. Your trading didn't. And you don't know why.
Here's what most traders do. You have an idea. Maybe you've noticed that when volatility spikes, it tends to come back down. So you build a strategy around it. Buy when VIX crosses above a threshold, sell when it drops below another. You run the backtest.
So you adjust. Maybe 28 and 20 is better. You try 23 and 16. You try thirty combinations. One of them produces a smooth equity curve with a strong Sharpe. You've found it.
You trade it. It falls apart.
The data was real. The backtest was honest. You didn't peek at out-of-sample. You did everything right. So what happened?
The scorecard and the bet
You optimized the scorecard. You never tested the bet.
And most people don't even realize those are two different things.
The scorecard is performance. Sharpe, drawdown, win rate. The bet is what you're actually predicting about the market. But here's the thing. If you're optimizing for Sharpe directly, wouldn't the system just find whatever produces great in-sample returns? Sure. But that doesn't tell you what's signal and what's noise. You can't tie a causal explanation to why the income exists. And if you can't explain why it works, you definitely can't explain when it stops.
So let's break this apart. Your strategy says "buy when VIX is above 25." But what are you actually predicting? You're predicting that VIX at that level means fear is overdone and the market will recover. That's the bet.
But you never tested the bet on its own. You never asked: when VIX is above 25, is fear actually overdone? Or is VIX high because it should be high? Because the market is about to get worse?
Those are completely different states. And your threshold can't tell them apart.
March 2020. VIX blows past 25. Your rules say buy. The market drops another 20%. VIX was high. But it wasn't overpricing fear. It was underpricing it. Realized vol was about to catch up. The bet was wrong. But you never tested it separately, so you had no way to know.
The bet needs a who
So what does a real bet look like? It's not "VIX is high, buy." It's not even "implied vol is above realized vol." Those are observations, not explanations. They describe what happened. They don't explain why.
And this is where most people stop. They find a pattern that correlates with returns and they call it a strategy. But correlation isn't a mechanism. So the real question is: where does the money come from?
Not "what pattern did I find." Who is on the other side? Why are they forced to be there? And what flow does that create?
Let's consider a specific example. Imagine a family office whose investment mandate requires them to hedge their long S&P 500 position when VIX is above 19. Every time VIX crosses that level, they buy 30-delta puts. That's a forced flow. A specific participant with a contractual obligation creating a predictable transaction.
You don't need to know which family office. You just need to know that forced hedging flow shows up in the data. Volume clusters at specific strikes. Open interest concentrates at specific tenors. So you can see the footprint without knowing who left it.
You know why it happens. You can predict when it shows up. And you can test whether the prediction is correct. Independent of whether trading on it makes money.
And the prediction isn't "the market goes up." It's "implied vol on this specific contract spikes in this window because of this flow." That's testable. That's specific enough to be wrong.
That independence is the whole point. You test the bet first. Does the forced flow actually happen when VIX crosses 19? Does option pricing actually spike temporarily? Does it reverse after? If yes, then you check whether trading on it is profitable. If no, you stop before wasting capital on a broken foundation.
But here's what's interesting. Even if you can confirm the mechanism exists, that doesn't mean the trade works. Because who else knows? If the maker is informed, spreads widen and there isn't enough of the pie for you to capitalize on. If other traders are in the same sequence, they're front-running the same flow. So the question shifts. It's not just "does the mechanism exist?" It's "who else is trading it, and where am I in the sequence?"
Most traders never get to any of this. They go straight from "I noticed a pattern" to "let me optimize until it backtests well." The who never enters the picture.
What happens when you don't have the bet
If the strategy stops working and you never separated the bet from the scorecard, you're stuck. You don't know if:
The prediction was wrong. You were never right about the market and the backtest found a coincidence.
The prediction is right but crowded. The mechanism exists, but too many people are trading it and the profit has been compressed.
The prediction is right but the execution is wrong. You're entering too late, sizing wrong, or trading in the wrong conditions.
Or the prediction was right, but the mechanism no longer exists. The regulation changed. The participant exited. The structural flow stopped. This is the one failure mode that only the bet can detect. If you know the who, you can monitor whether they're still in the market. Without the bet, a regime change and a wrong prediction look identical.
These have completely different solutions. But without the bet, they all look the same: "it stopped making money." So you throw the whole thing out and start over.
Though here's what's worse. Sometimes you throw out a strategy where the bet was actually right.
The pie
Think of the edge as a pie. The pie is bounded by the notional size of the structural flow. First person to trade it gets the biggest slice. Second person gets less. By the time hundreds of traders are doing the same thing, each slice is smaller than the transaction cost. The edge is real. The profit is gone.
So if your scorecard is returns and returns went to zero, you conclude the signal is dead. You delete it and move on. But the mechanism never changed. The prediction is still correct. The forced flow still happens. The information the signal carries is still valuable. Who's trading, why, and when. It just answers a different question than the one you were asking.
Most of the edge traders abandon isn't dead. It's pointed at the wrong target.
And this gets at something deeper. Edge isn't a trade. It's knowing that a mechanism exists, why it exists, and whether it's still active. The trade is what you do with that knowledge. Sometimes the trade is to enter. Sometimes it's to sit out. Sometimes it's to use the information for a completely different purpose. But the knowledge itself is the asset. And it compounds. Because every mechanism you've mapped teaches you where to look for the next one.
The step
This isn't about VIX. It's about every strategy you've ever built. Whatever market you trade, the step is the same.
Before you optimize anything. Before you touch a parameter, run a backtest, or look at a single equity curve. Define the bet.
What specifically are you predicting about the market? Not "it will go up." Not "volatility will come down." What mechanism are you claiming exists? Who is the structural participant? What would be true if you're right? What would be true if you're wrong? Can you test the prediction without looking at returns?
If you can't answer these, you don't have a strategy. You have a scorecard with no bet behind it. And it will work until it doesn't, and when it stops, you won't know why.
And if you already test the bet first, the question becomes: are you testing every possible signal, or just the ones you can imagine?
That's the step.