Research methodology

How to find stock market alpha

Alpha is return in excess of a benchmark over a stated horizon, attributable to a mechanism you specified in advance. Almost everything that fails in practice fails at the specification step: no benchmark, no horizon, a one-step story, and a sample that quietly excludes the times it did not work. What follows is the diagnostic, applied to one ticker at a time, with a worked success and a worked failure from our own published ledger.

As of 2026-07-27 10:55Z·Methodology mo-seo-2026.07·Reviewed by Market Ontology Research

The diagnostic

  1. 01

    Fix the benchmark before the idea

    Alpha is a residual, so it does not exist until you name what it is a residual to. For a single US large-cap that is usually SPY or a sector ETF; for a cyclical it is often the sector, because beating the index on an oil rally is not evidence about the company. Write the benchmark down first, because choosing it after the fact is the most common way a research process fools itself.

  2. 02

    Fix the horizon before the idea

    A mechanism that works over one quarter and a mechanism that works over three days are different claims requiring different evidence. State the horizon in trading sessions. If you cannot say when the thesis should have resolved, you cannot say later whether it failed.

  3. 03

    List the events that plausibly move this name

    Scheduled events first: earnings, guidance, index rebalances, macro releases the name is sensitive to. Then unscheduled: policy decisions, supply disruptions, litigation, competitor actions. The point is coverage — an event you have not listed is an event you will rationalise after it happens.

  4. 04

    Write the causal path in at least two steps

    Event changes a variable; that variable enters revenue, margin, capital intensity, or the discount rate; that changes the value. If your path is one step — 'sanctions are bullish for defense' — you have a slogan, not a mechanism. Two steps forces you to name the quantity that has to move.

  5. 05

    Measure the observed response, including the failures

    For each past instance of the event, compute the return over your horizon minus the benchmark return over the identical window. Keep the instances that went against you in the sample. A mechanism that worked four times out of five is informative; a mechanism where you only remember the four is not.

  6. 06

    Check whether a factor already explains it

    Compare the residual to simple factor exposures — size, value, momentum, and the sector itself. If momentum explains the abnormal return, you have found a factor tilt you can buy more cheaply elsewhere, not alpha.

  7. 07

    Name the next evidence date

    The date on which new information will either support or contradict the thesis. Without it, a position drifts into a holding, and a holding has no falsification date.

Worked examples from the published ledger

Currently unavailable — no forecast horizon has closed and been measured yet, so no worked success or failure can be shown honestly
Locked, awaiting measurement

IPG Photonics to Acquire Lumibird Medical for €300 Million → Bullish IPGP

IPGP · locked 2026-07-27 · benchmark SPY · horizon 1–4 weeks · direction bullish

This is what step seven looks like in practice: the claim is on the record with its benchmark and horizon before the outcome is known.

Run it on your own ticker

The diagnostic is mechanical enough to automate: fix the benchmark and horizon, pull the relevant events, write the two-step path, measure the residual across every past instance, and check it against factor exposure.

Run this diagnostic on my ticker

Related