For almost a century, the so-called Presidential Predictor, as dubbed by Sam Stovall—a well-known S&P analyst and the chief investment strategist for CFRA—has correctly determined the presidential election winner with only one exception, an 88% success rate.

It works like this: If the S&P 500 rises from August through October, the incumbent party (in this case, Republican—Trump) is reelected to the White House. If the S&P 500 falls over the three-month stretch, the Oval Office flips to the opposing party (in this case, Democrat—Joe Biden).

This year, for much of that period, stocks have soared to new highs, predicting a Trump reelection; as recently as mid-October, the S&P 500 Predictor was up 7.5%.

But that changed Wednesday as stocks plunged further. Now the S&P 500 is down over the critical time frame, albeit ever so slightly—by just 0.09 (such a small difference that the percentage change is basically flat), to 3,271.03.

That number will be the level to watch over the next couple of days: Because the Presidential Predictor counts the stock market’s performance only through October, Friday is the last day that matters for the indicator. Should the S&P 500 remain where it is or fall further over the rest of this week, the stock market predicts that Biden will become President.

“The next two days will be key!” Stovall noted in an email.

Stock market analysts believe the reason for the pattern is that investors dislike uncertainty: A change in control of the White House comes with uncertainty around new policies and legislation. The more likely that change becomes, the more uncertainty there is, which makes investors sell stocks.

Still, the metric isn’t perfect. In 1956, despite the S&P 500’s fall in the August-through-October period, Dwight D. Eisenhower won reelection to the presidency. The Presidential Predictor has not failed since then.