Why Only Momentum and Quality Remain — Series Wrap-Up

This series started from a single question: can you find a market-beating signal in public data? The table below is the result of testing dozens of hypotheses with the same method (stated sample, benchmarked against the market, costs included, look-ahead blocked).

Overall Test Results

And what we kept in a live account was not a newly discovered signal, but the two factors academia documented decades ago — momentum (3/6/12-month returns) and quality (gross profit/assets, revenue growth).

Why These Two

We did not discover them. Only what had already been validated by a century-scale out-of-sample record also passed our tests. Momentum is the most robust anomaly confirmed in the academic literature across nearly two centuries of US market data and across many countries and asset classes (extensively re-tested since Jegadeesh & Titman 1993), and quality is the phenomenon that gross profitability predicts subsequent returns (Novy-Marx 2013).

Surviving factors share one condition. Even when you know about them, they must be hard to hold. Momentum crashes periodically — we paid that price in July 2026 with -32.5% in a single month. If the signal were easy to hold, arbitrage would have eliminated it long ago. The premium remains not as compensation for information but as compensation for pain.

Why the Search for New Factors Fails

There are three layers of wall. ① Price, financial and investor-flow data have already been ground through by quants worldwide — most of what you can find on a laptop at a monthly frequency is already in prices. ② In monthly returns, noise is dozens of times the signal, so even real alpha cannot be distinguished from luck without decades of data. That is why running many backtests inevitably "finds" something fake (the backtest pitfalls piece). ③ Even factors that were real die by roughly half once they become known (post-publication decay is a repeatedly confirmed finding in the literature).

What Is Left for Individual Investors

The structural advantages these tests point to lie outside factor space:

1. Size — Narrow opportunities that institutions cannot enter (a signal worth 2.4 days a year) survive because they are small. An individual can enter precisely because they are small
2. Patience — An individual with no career risk can hold through a momentum crash. If the premium is compensation for pain, a structure that can endure pain is itself an edge
3. Events — Not cross-sectional factors, but short-half-life disclosure events still have something testable left. (Correction 2026-09-04: the insider cluster example originally cited here was rejected after a look-ahead error was found — Part 17. Its half-life was under one day)

The cheapest conclusion goes last. The fact that "it doesn't work" came up this often across dozens of hypotheses is itself the core output of this series. Knowing what does not work, from data, reduces the cost of wandering in search of what does — one of the few advantages an individual can reliably hold.

Data Sources

• Verification data from the entire series (Korea Exchange daily prices and investor flows, SEC filings, our own trading records)