Small Survivors — Two Narrow Signals That Survived 16 Years of Testing

Most of the signals this series tested on the Korean market were rejected. Two escaped rejection — and both survived not because they are strong, but because they are narrow.

Survivor 1 — Buying Down Days During Extreme KOSPI Slumps

• Signal: buy the index on a down day while KOSPI sits 5% or more below its 20-day moving average (disparity ratio below 95)
• Sample: over 16 years there were 94 such days — an average of 2.4 days a year
• Result: next-day average +0.51%, win rate 60% (significantly higher than the baseline across all down days)

Dip buying on individual stocks was rejected in Article 8, but extreme slumps at the index level were different. Interpretation: a sharp drop in an individual stock is often information (bad news about that company), while an extreme index slump is consistent with a regime in which forced liquidation and panic weigh more heavily than information.

The limits are recorded as they are: grouped into episodes, those 94 days amount to 4–6 regimes. Statistically that means "it worked across 4–6 crises in 16 years," and the assurance that it will work in the next crisis is only as strong as that sample count.

Survivor 2 — The Rights-Offering New-Share Listing Calendar

• Observation: for stocks issuing rights offerings or CBs, the listing date of the new shares is a calendar disclosed in advance. Stocks that spiked (+15% or more) within the two weeks before listing tended to keep running after listing as well
• Sample: 403 cases
• Result: +4.5% over the 20 days after listing, win rate 56% — but dispersion of ±40%. The extreme cases where the preceding spike exceeded +80% scattered from -43% to +105%

A 56% win rate with ±40% dispersion means "a coin flip slightly in your favor on average, but a very large coin." The verdict was left as undetermined, not valid.

What the Two Survivors Have in Common

• Capacity is small. You cannot run an account on a signal that fires 2.4 days a year and a calendar with ±40% dispersion. This is most likely the very reason institutions have not arbitraged these signals away — there is too little to take
• There is a structural reason. Panic liquidation (Survivor 1), the incentive to prop up the price before listing (Survivor 2) — only signals with a mechanism hypothesis behind the numbers have grounds to work in the next sample

The fact that dozens of tests left only these two is itself the introduction to the final article in this series, Why It Comes Down to Momentum and Quality.

Data Sources

• Korea Exchange daily quotes (indices and individual stocks) and corporate disclosures. Long-term index trends can be checked on the market page