There's a line that surfaces every time panic breaks out somewhere: stocks have recovered from every war. The line is true, and the numbers behind it are correctly calculated.
They describe one country.
What happened in the United States
The Second World War. German forces entered Poland on 1 September 1939. Over the following fortnight the Dow rose from 131 to 155, about eighteen percent, including a single day's gain of 7.3 percent. Wall Street was pricing European demand for American goods. Brokerage houses actively promoted how well the so-called "war bride" stocks had performed after 1914.
The buying panic was over by late September. That peak was not seen again until close to the end of the war.
The real pre-war high came on 8 April 1940 at 151.29. From there it fell — through the invasion of Western Europe, through Pearl Harbor and a drop of roughly three percent in a day — to 28 April 1942, when the Dow stood at 92.92. That is just under thirty-nine percent below the high.
From there the recovery ran, and it was impressive: up eighty-seven percent to the end of the war in August 1945. From Pearl Harbor through December 1945 that works out at 13.6 percent a year.
One number appears in none of those charts. The Dow did not close above 151.29 again until 29 December 1944. Four years and almost nine months under water. Anyone who bought on 8 April 1940 spent that entire stretch without a single day in profit.
That is the figure that decides whether someone holds a position — and it's missing from nearly every presentation, because those show the depth and never the duration.
The First World War. Here there's a more fundamental problem. The New York Stock Exchange closed on 31 July 1914 for roughly four and a half months — the longest closure in its history. Stocks lost about thirty percent at the outbreak, and then there was simply no price for months.
A position you cannot sell has no drawdown. It has a closed market, which is a different thing.
After reopening, 1915 delivered a gain of more than eighty percent, the best year in the Dow's history. Roughly thirty-two percent was given back across 1916 and 1917. Then came the twenties.
All of these are price returns, excluding dividends. With substantial wartime inflation, the real figures are considerably lower.
The other half of the world
Both stories above concern a country that won both wars and on whose mainland neither was fought.
Elsewhere:
Russia. The St Petersburg exchange rose one hundred and fifty-five percent between 1908 and 1914. Then came the war, the revolution in 1917, and expropriation. Foreign shareholders received nothing. Not little — nothing. Even after 1990, original shareholders were never recognised or compensated.
Austria-Hungary. The state whose companies were listed there ceased to exist.
Germany, First World War. The market survived in nominal terms. In real terms, the hyperinflation of 1923 destroyed whatever was left.
Japan, Second World War. In the Dimson-Marsh-Staunton database, the Japanese 60/40 portfolio records a maximum drawdown of almost ninety-eight percent — the deepest of any market covered. The Tokyo exchange closed in August 1945 and did not reopen until 1949.
China, 1949. The Shanghai exchange was nationalised. Original shareholders were never compensated.
Why those markets appear in no statistics
Here is the actual point.
The datasets from which "stocks recover from every war" is assembled typically contain sixteen markets with continuous series. Sixteen markets with continuous series are sixteen survivors.
A destroyed exchange produces no continuing series. It simply stops. And what stops drops out of the analysis — not through carelessness, but because there are no further numbers to include.
Argentina was among the largest equity markets in the world around 1900. It appears in none of the standard long-run studies.
The effect has two names, describing two halves of the same problem. Survivorship bias is the familiar one: you analyse what is left. Easy data bias is the less familiar and perhaps more important one: you use the data that is easy to obtain. And American data is easy to obtain because America did well.
What emerges when you do it differently
Aizhan Anarkulova, Scott Cederburg and Michael O'Doherty built a different dataset for exactly this reason: thirty-nine developed markets, 1841 to 2019, roughly two thousand seven hundred country-years. Explicitly to sidestep survivorship and easy data bias.
Their result, published in 2022 in the Journal of Financial Economics: the probability that a diversified investor is down in real terms after thirty years is twelve percent.
In the follow-up work, thirty-eight countries from 1890 to 2019, the numbers run: domestic stocks thirteen percent loss probability over thirty years, bonds twenty-seven, bills thirty-seven. And the losses do not arrive neatly separated — simultaneous real losses across several asset classes over long horizons are not rare.
The authors note themselves that this contradicts the conventional advice that stocks are safe over long holding periods.
Even that figure is the friendly version. A dataset of thirty-nine developed markets still contains only countries that remained countable.
Why this is here
Because it is the same problem we deal with daily.
Test a crypto strategy on today's hundred largest coins and you are testing it on survivors. Everything that died between 2018 and 2024 is missing — and a trend filter looks better on survivors than it was. Same mechanism as the sentence about the wars, over six years instead of a hundred.
The difference is that we can fix it for coins as soon as historical rankings exist. For countries, nobody can. The St Petersburg exchange has no post-1917 series because the exchange no longer existed.
What remains
Three sentences, all true, which have to be read together:
The American stock market came through both world wars and rose substantially afterwards.
Anyone buying in April 1940 needed four years and nine months to get back to even — and in the summer of 1914 could do nothing at all for four and a half months.
And for investors in St Petersburg, Vienna, Tokyo and Shanghai the first sentence isn't true, because they are no longer in the statistics.
Which of those three you quote isn't determined by the evidence. It's determined by which dataset you opened.
Not investment advice, not a forecast. All index figures are price returns, excluding dividends.
Sources: Anarkulova, Cederburg & O'Doherty, "Stocks for the long run? Evidence from a broad sample of developed markets", Journal of Financial Economics 143(1), 2022, pp. 409–433, and the same authors on long-horizon returns of stocks, bonds and bills (38 countries, 1890–2019) · Dimson, Marsh & Staunton, "Triumph of the Optimists" and the continuing DMS database · historical Dow Jones price data and documented New York Stock Exchange closures
Study the Past — Improve your Future 🥋