Hindsight Bias in Investing: Why the Past Always Looks Easy

What hindsight bias is, and where the idea came from

In 1975 Fischhoff, then at the Hebrew University of Jerusalem, gave students a 150-word account of a little-known episode, such as the British campaign against the Gurkhas of Nepal that began in November 1814. Every version listed four possible outcomes. One group read the passage with no ending. Four other groups read the same text with one extra sentence stating which outcome had "really" happened, and three of those endings were false.

Whatever ending people were given, they rated it as more likely than the uninformed group did, and they judged the details that supported it as more relevant. They also could not see the effect in themselves: in a follow-up experiment, people told to answer as they would have without knowing the outcome still leaned toward the ending they had been given. Quoting the historian Georges Florovsky, Fischhoff named the tendency creeping determinism, the sense that what happened could not have happened otherwise.

A companion study by Fischhoff and Ruth Beyth asked people to estimate the odds of various events before Richard Nixon's 1972 trips to Beijing and Moscow. Two weeks to six months later the same people were asked to recall their forecasts. Their remembered odds drifted toward what they believed had actually happened. Later researchers Neal Roese and Kathleen Vohs split the bias into three layers: misremembering what you predicted, feeling that the outcome was inevitable, and believing you personally could have foreseen it.

Markets are an ideal breeding ground for all three layers. Prices leave a permanent, public record, and the record arrives with the ending already attached.

Every crash looks obvious afterwards

The classic hindsight bias examples in investing are crashes. After 2000 everyone remembers the profitless dot-coms; after 2008, the subprime loans. What memory edits out is how often the same kind of warning appeared in years that ended fine. A sharp drop is the most common "signal", so we counted how often it happened.

Using the S&P 500 price index (dividends excluded) from 1963 through 2025, the index fell at least 10% below its year-to-date high at some point in 34 of 63 calendar years. In 19 of those 34 years it still finished the year higher. Slides of 20% or more within a calendar year happened in 11 years, including 1987, 2008, 2020 and 2022. In real time, an investor sees a 10% drop with no label telling them whether it is 1987, which ended the year up, or 2008, which did not.

That is the base rate hindsight erases. The 2008 crash and the dot-com bust each had loud warnings, but so did plenty of years when the warnings came to nothing.

Great compounders looked broken while it was happening

The mirror image of the "obvious crash" is the "obvious winner". Today Amazon, Apple and Nvidia look like inevitable successes. Each of them went through a stretch when holding the stock felt like a mistake, and the news at the time gave holders good reasons to sell.

The stories behind those numbers did not sound like buying opportunities. In June 2000 a Lehman Brothers convertible-bond analyst, Ravi Suria, argued that Amazon could run out of cash within four quarters unless it raised money or cut spending; the shares fell 19% on June 23, 2000, and kept falling for another 15 months. Apple bottomed on January 20, 2009, less than a week after Steve Jobs announced a medical leave in the middle of the financial crisis. In 2011 Netflix split its DVD and streaming plans into separately priced subscriptions, a steep increase for customers who wanted both, announced and then abandoned the Qwikster spin-off, and reported in October that it had lost roughly 800,000 US subscribers. The stock fell 35% the next day. Meta lost 26% on February 3, 2022, its worst day on record at the time, after reporting the first quarterly drop in Facebook's daily users.

Now look at the last row. Cisco's decline looked just like Amazon's while it was happening, and the narrative was arguably better: a profitable market leader selling the equipment the internet ran on. Even with dividends reinvested, it took until August 2021 to get back to its March 2000 peak. Hindsight keeps the rows that recovered and forgets the ones that did not, which is also how survivorship bias distorts the picture.

The blind chart exercise

Here are two real large-cap technology stocks over the same five years, normalized to 100 on December 1, 1999. You can see the dates. You cannot see the names. Before reading further, decide which one you would have wanted to own at the end of 2004, and why.

What you had to go on: Asset B rallied to 175 in March 2000 before collapsing to 19 by October 2002. Asset A did worse. It slid from a peak of 126 within days of the start to 7 in September 2001, a 94% loss, and finished 2004 at 52. Asset B finished at 42. For comparison, the S&P 500 price index ended the same window 13% lower. On this evidence the two look similar: both badly damaged, both rebounding in 2003, both slipping again in 2004. Asset A had the deeper wound.

The reveal

Asset A is Amazon. Asset B is Cisco Systems. From the last day of 2004 to October 1, 2026, Amazon's shares multiplied about 112 times. Cisco, with dividends reinvested, multiplied almost 9 times. Similar scars at the end of 2004, a roughly thirteenfold difference in what came next.

Now notice what happens in your head. Once you know the names, the first chart starts to look different: Amazon's 2003 rebound reads as a turning point, Cisco's 2004 slide as a warning. Neither reading was available in 2004. That shift is exactly what Fischhoff's students experienced with the Gurkha passage.

The dart-throwing monkey and the WSJ Dartboard Contest

If skilled forecasting is hard to tell apart from luck, a random picker should do about as well as an expert. That was Burton Malkiel's provocation in his 1973 book A Random Walk Down Wall Street, where he wrote that "a blindfolded monkey throwing darts at a newspaper's financial pages" could match a portfolio chosen by experts.

The Wall Street Journal turned the dart-throwing monkey into a column. Starting in October 1988, with Malkiel throwing the first dart, investment professionals each picked one stock while Journal staffers threw darts at the stock tables, and both were measured against the Dow Jones Industrial Average. Contests ran one month at first and six months from 1990. When the Journal announced in April 2002 that it was retiring the feature, the tally over 142 six-month contests since 1990 was:

• Pros beat the darts 87 times to 55.
• Pros beat the Dow 76 times to 66.
• Average six-month gain: pros 10.2%, darts 3.5%, Dow 5.6%.

That looks like a clear win for the pros, and on raw numbers it was. The fine print matters. Returns were price-only, dividends excluded, and the pros favored lower-dividend stocks with slightly higher betas, which Malkiel noted tend to rise more in a rising market. The picks were also published in one of the most widely read newspapers in the country. Brad Barber and Douglas Loeffler (1993) found an abnormal gain of about 4% in the two days after publication that partly reversed within 25 days. Bing Liang (1999), studying the 1990-1994 contests, concluded that once the publicity effect, the missing dividends and the small sample were accounted for, the pros beat neither the market nor the darts. Malkiel himself said four darts were never what he recommended: his point was owning a broad cross-section of the market.

Random-portfolio studies

Later researchers replaced the darts with computers. Rob Arnott and colleagues (2013) built random and deliberately naive portfolios from the 1,000 largest US stocks for 1964-2012 and found they generally beat a capitalization-weighted index before trading costs. A Cass Business School team (Andrew Clare, Nick Motson and Stephen Thomas, 2013) generated 10 million randomly weighted indices of the 1,000 largest US stocks for 1968-2011, and almost all of them beat the cap-weighted version.

This is where monkey stock picking gets misread. Random weights put more money into smaller and cheaper companies than a cap-weighted index does, and Arnott's team found that exposure to those two well-known factors, size and value, explained most of the extra return. The monkeys did not know anything. The studies say a great deal about how indices are built and very little about whether to trust a stock picker. For the evidence on stock picking vs index funds and how often professionals beat the S&P 500, see our guide to beating the market.

How to fight hindsight bias in practice

Research is fairly blunt that you cannot simply decide to ignore what you know; Fischhoff's subjects were told to and failed. What works is changing the process so the pre-outcome view is preserved and challenged.

1. Keep a decision journal. Before any purchase or sale, write down what you expect, why, how confident you are as a percentage, and what evidence would prove you wrong. Date it. Fischhoff and Beyth showed that memory quietly rewrites forecasts; a written record does not.
2. Run a pre-mortem. Gary Klein's method, described in Harvard Business Review in 2007, asks you to imagine the decision has already failed and list the reasons. Klein cites 1989 research by Deborah Mitchell, Jay Russo and Nancy Pennington that found this kind of prospective hindsight improved people's ability to correctly identify reasons for future outcomes by 30%.
3. Start from base rates. Before reading a drop as a signal, ask how often a similar drop led nowhere. For the S&P 500, a 10% slide inside a calendar year has been more common than not since 1963.
4. Consider the opposite. Roese and Vohs point to deliberately working out how a different outcome could have happened. Applied to the table above: write the bull case for Cisco in 2002 and the bear case for Amazon, using only what was known then.
5. Grade decisions separately from outcomes. A sound decision can lose money and a reckless one can win. Review your journal entries for the quality of the reasoning, then look at the result.

Why Blindfolio hides the names (and, in random mode, the years)

Hindsight needs two clues: the name of the company and the date. Blindfolio always removes the first, and in random mode the second as well. A round covers a 10-, 15- or 20-year stretch somewhere between 1962 and 2026, and companies appear under temporary codenames throughout. In a random round the year is hidden too: it is masked in the date and in the historical news headlines you see along the way, and the era is revealed only at the end. If you pick a start decade instead, you see the dates, but the names stay hidden. Either way, the real names appear only at the end, next to a comparison with the S&P 500, dividends reinvested, over the same period. The catalog includes companies that later went bankrupt, such as Enron, Lehman Brothers and Pets.com, so the pool is not filtered for survivors. More on how a round works is on the stock market game page.

This reduces hindsight rather than removing it. An experienced player can sometimes guess the era from a famous crash shape or the level of interest rates. Even so, making decisions with no names while the clock runs week by week is a fair approximation of what Fischhoff's "Before" group faced: the information that was available at the time, and no ending.