Framing a Falsifiable Directional Thesis
Objectives
- State a directional thesis as a falsifiable claim with an explicit asset, direction, horizon, magnitude band, and a pre-committed kill condition.
- Convert the five-gate honest read into the design checklist a thesis must survive before it is worth testing — scale, shape, spread, base rate, and uncertainty.
- Specify, in advance, the single observation that would force you to abandon the thesis, and explain why pre-registering it is what makes the claim scientific rather than narrative.
Everything you have built in this institute has prepared you to do one honest thing: decide whether a number is genuine evidence that an asset will move. The capstone asks you to turn that judgment around and aim it at yourself. You will state a thesis — a directional claim about a real asset — and then you will try, in good faith, to destroy it. The thesis that survives an honest attempt to kill it is the only kind worth trading. The thesis that cannot even be killed is not a thesis at all. So before you touch the formula engine in the next chapter, you have to learn to frame a claim that is capable of being wrong.
Start with the intuition, because it is the whole discipline in one idea. A belief that no possible observation could contradict is not knowledge — it is faith with a chart attached. "This asset is bullish" is unfalsifiable: bullish over what horizon, by how much, and what exactly would you have to see to admit you were wrong? If there is no answer to that last question, the claim can absorb any outcome. It went up, so you were right. It went down, so it is "still early." A thesis that explains every outcome predicts none of them. The philosopher Karl Popper gave us the test that quant research lives by: a statement earns the right to be called scientific only when it forbids something — only when there exists a clear observation that, if you saw it, would prove the statement false. Your job in this chapter is to write a directional claim that forbids something specific.
So let us formalize what a falsifiable directional thesis must contain. It has five parts, and you will notice they are not new — they are the five gates of the honest read, turned from a reading checklist into a writing checklist.
The first part is the claim itself, stated in the right scale. Name the asset, the direction, the horizon, and a magnitude band. Not "oil goes up," but "the log-return of crude over the next twenty trading days is positive and at least three percent." Putting it in log space is not pedantry; it is the first gate. A claim framed in raw arithmetic space silently changes meaning with the starting price, and a thesis whose meaning drifts cannot be falsified cleanly. State the move in the additive, symmetric space where a plus-three and a minus-three are true opposites.
The second part forces you through the shape gate. Your directional claim is a statement about the center of a distribution, but the outcome that matters often lives in the tail. Ask what the distribution of this asset's twenty-day returns actually looks like. If it is heavily skewed or fat-tailed, a thesis that only names a median direction is hiding most of its own risk. A well-framed thesis says not just where it expects the asset to go, but what the shape of being-wrong looks like — because that shape is where the loss lives.
The third part is the spread gate, and at the thesis stage it sets your bar for surprise. A three-percent move means something completely different in a calm regime than in a turbulent one. If twenty-day volatility is running at two percent, a three-percent drift is a real signal; if volatility is running at eight percent, the same three percent is noise the market produces by breathing. Frame your magnitude relative to the prevailing spread, so the claim does not quietly become trivial the moment the regime changes.
The fourth part is the base rate, and it is the gate that kills the most theses before they cost anyone money. Ask the deflating question: what was going to happen anyway? If this asset closes higher over twenty days fifty-five percent of the time regardless of your reasoning, then a thesis predicting "higher" has to beat fifty-five percent to be worth anything — not beat fifty. Your thesis is evidence only to the extent it improves on the background rate. Write the base rate down before you fall in love with the idea, because a pattern that merely matches it is noise that has learned to dress well.
The fifth part is the uncertainty gate, and here it becomes the kill condition — the single most important sentence you will write. Every thesis is an estimate from a finite sample, surrounded by an error bar, and you must decide in advance what observation would push the truth outside that bar. Pre-commit to it: "I abandon this thesis if the asset closes below this level," or "if the directional gap collapses inside its noise band for three straight weeks." The kill condition has to be written before you see the outcome, because a stop you move after the fact is not a test — it is a rationalization with a timestamp. The number you write down here, while you can still be honest, is what converts a story into science.
Notice what you have done. You have not yet tested anything. You have simply refused to let yourself hold a claim that cannot be wrong. That refusal — taking a belief you find attractive and bolting a kill condition onto it before you risk a dollar — is the entire spirit of the capstone, and it is the difference between a track record you can defend and a collection of lucky narratives.
Take a live directional thesis into Saqrid and run it through the five-gate honest read, the same checklist you learned in Foundations — but now you are vetting your own claim rather than the platform's output. Suppose your thesis is "crude is bullish over the next month." Gate one, scale: restate it in log space as a positive twenty-day log-return of at least three percent. Gate two, shape: pull the distribution of crude's twenty-day returns and check whether a left tail dwarfs your upside — if it does, your bullish median is sitting on a fat downside. Gate three, spread: read current twenty-day volatility; if it is eight percent, your three-percent target is inside the noise and the thesis is empty as written. Gate four, base rate: Saqrid tells you crude closed higher over twenty days about fifty-four percent of the time in this regime, so "higher" alone clears nothing — your edge has to beat fifty-four. Gate five, uncertainty: pre-register the kill — "abandon if crude prints a twenty-day log-return below minus two percent, or if the signal sits inside its surrogate band for three weeks." A thesis that survives all five gates as a written claim is one you have earned the right to backtest next chapter. One that fails even one gate is not a thesis — it is a hypothesis you have not yet earned the right to believe.
Frame this in TEPLab as your candidate capstone thesis: "Over the next twenty trading days, the asset's log-return is positive and at least one standard deviation of its trailing twenty-day return distribution, and this beats the regime base rate by at least eight percentage points." Write it down with an explicit, pre-registered kill condition — for example, abandon if the realized twenty-day log-return falls below minus one standard deviation, or if the live edge collapses inside its bootstrap confidence band. You will not backtest it yet; the deliverable for this chapter is the falsifiable claim itself, framed so cleanly that next chapter's engine can only confirm it or kill it — never quietly excuse it.