Nonzero Lab Research · Statistics · 2026-09-14 · 7 min read

Catching a Falling Knife: How Often Do Losers Come Back?

Intel took 26 years to reclaim its 2000 high and set a record in June 2026. But of 25,967 US stocks since 1926, only 42.6% beat T-bills over their lifetime.

What actually happened on Intel's chart?

Set the narrative aside and look only at price.

On August 2, 2024, Intel closed down 26.1%, from $29.05 to $21.48. On April 8, 2025 it closed at $18.13, its lowest close of those two years and 75.8% below the 2000 peak.¹ That August 22, the US government announced an $8.9 billion purchase of 433.3 million primary shares at $20.47 per share, equivalent to a 9.9% stake, funded by $5.7 billion in previously awarded but unpaid CHIPS Act grants plus $3.2 billion from the Secure Enclave program.³

On April 24, 2026, Intel rose 23.6% in a single session to close at $82.54 — its first close above the 2000 high since 2001. Two months later, on June 22, it closed at $140.94, up 677% from the April 2025 low. The most recent session, September 11, 2026, closed at $102.94, 27% below the June peak.¹

Now cut that chart at April 8, 2025 and cover everything to the right. What you see is a line down three quarters from its all-time high, fresh off the second-worst single day in company history, dividend suspended, layoffs underway. No bar on that chart tells you a 677% move is coming. Cut it at June 22, 2026 instead, and no bar tells you a 27% drawdown is coming either.

That is the whole problem. This sequence reads like a coherent story only because we are standing at the end of it, looking back.

How often do beaten-down charts come back?

To answer that you have to leave the anecdote and look at the distribution.

Bessembinder (2018) measured lifetime buy-and-hold returns for the 25,967 common stocks listed in the United States between July 1926 and December 2016, and found two things. First, only 42.6% of individual stocks beat one-month Treasury bills over their lifetime — more than half of all US stocks, held to delisting or to the end of the sample, did worse than T-bills. Second, the entire net wealth creation of the US stock market since 1926 came from roughly the best-performing 4% of companies; every other stock combined returned about what Treasury bills did.²

Read that on its own terms. It measures a stock's whole life, from listing to delisting or the end of the sample — not "the odds of recovering from a given drawdown." It does not separate "rose then fell" from "never rose at all." Strictly speaking it is not a direct answer to "what fraction of stocks down 76% come back"; that number requires its own study, and this article does not supply one.

What it does establish is the shape of the distribution, and the shape is half the answer. Single-stock returns are extremely right-skewed: the median stock badly lags the index, and nearly all of the gain sits in a tiny minority of names. When a distribution looks like that, the prior is not on the side of "this one will come back."

It also explains why reasoning from index history to single-stock dip buying is a category error. An index reconstitutes — failures get removed and replaced. It is, by construction, a portfolio that continuously drops its losers. Your single position does not do that for you.

The literature does not agree on buying losers

If the base rate does not support it, does the pattern? Here the evidence splits, and splits cleanly.

De Bondt and Thaler (1985) sorted stocks on three-year past performance and found that the "loser" portfolio significantly outperformed the "winner" portfolio over the following three years, which they read as overreaction to bad news.⁴ Eight years later, Jegadeesh and Titman (1993) ran the same exercise with three- to twelve-month ranking periods and got the opposite result: past winners kept beating past losers over the next three to twelve months, with the long-short portfolio returning close to 1.5% a month.⁵

Both papers have been cited tens of thousands of times. Neither has been overturned. The difference is mostly horizon: momentum over three to twelve months, reversal over three to five years. They are not describing the same thing.

For someone reading a chart, the implication is direct. "It's fallen too far" and "weak stays weak" each hold in a documented regime, and the segment on your screen will not tell you which regime it belongs to. Treating "it's already down this much" as a reason to buy means picking whichever of two opposed findings feels better right now. That is a choice, not evidence.

Why does the chart look so reasonable in hindsight?

Looking back at Intel, everything connects: the government stake provided a financial backstop, advanced process nodes reached volume, foundry negotiations advanced, and the stock recovered. Every link has a place on the timeline.

The trouble is that the timeline was assembled backwards from the outcome. Fischhoff (1975) showed that once people are told how something turned out, they cannot reconstruct the judgment they would have made without that knowledge — not even when explicitly instructed to pretend they do not know.⁶ Anyone staring at that chart in April 2025 could not see the earnings report a year out, and did not know a $140.94 print was coming in June. They saw a line that had been falling for a long time, next to news that did not look like it was about to improve. Why Chart Review Lies to You: Hindsight Bias and Blind Practice works through that mechanism in detail.

There is a second, harder-to-notice bias stacked on top. You are reading an article about Intel precisely because Intel came back. The companies that did not come back do not get articles written about them; they delist, get acquired, or sit quietly at the bottom until nobody mentions them. The news stream is itself a survivorship filter, and most people's intuition about "deep dips bounce" was fed by that filtered sample.

Together the two biases produce a specific distortion: in hindsight, recoveries look both predictable and common. In foresight they are neither.