According to the bets placed on the race—which often predict outcomes more accurately than polls—Republican candidates David Perdue and Kelly Loeffler were both cruising to relatively easy wins. “Until just before the election, people betting their own money reckoned that Republicans had this in the bag,” says Thomas Miller, a Northwestern University data scientist who calculated the chances of victory in each race based on prices from the PredictIt.org political betting site. (You can view Miller’s data at https://twitter.com/virtualtout.)

Georgia may indeed be gradually turning blue, or at least purple. But until days before Jan. 5, its voters appeared to be upholding the rock-ribbed tradition of backing Republicans in statewide races, and especially in runoffs. The sudden drop in Loeffler’s and Perdue’s fortunes must rank among the sharpest declines, over the briefest periods, in the annals of major national elections. So how did the most important senatorial contests in a generation shift in a couple of days from bright red to medium blue in Miller’s betting odds, which indeed nailed the final tally?

As of the morning of Jan. 7, the final vote showed Democrat Raphael Warnock defeating incumbent Sen. Kelly Loeffler by 50.8% to 49.2%, and Democratic challenger Jon Ossoff beating Sen. David Perdue, who was seeking a second term, by a narrower 50.4% to 49.6%, with 100% of the ballots counted. The Ossoff-Warnock sweep empowers the Democrats to take charge in the Senate and control both houses of Congress. On the world’s betting sites, that outcome was a long shot offering two-to-one odds a big payout as recently as two days after Christmas.

Political experts will be studying the reasons for this stunning reversal for years to come. The factors are many, and difficult to untangle. According to Georgia political experts speaking to Fortune, including GOP strategists, the Republicans fell from favorites to underdogs for two major reasons. The first is simply the trend in the early vote. Between Christmas and New Year’s Day, bettors who were studying the early trends realized that the turnout was both gigantic and tilting more toward Democrats than in the presidential race.

Second, a GOP victory hinged on its leadership’s focusing relentlessly on the Georgia races. President Trump’s impact on the contest is hard to measure. But it’s clear that instead of crusading for Loeffler and Perdue, Trump threw his weight behind a smattering of issues that divided rather than unified the party: from claiming fraud in the Nov. 3 election to threatening to veto the GOP-backed stimulus package to calling Georgia’s Republican Secretary of State to demand that he “find the votes” to overturn Trump’s loss there. To paraphrase one veteran Georgia pol’s assessment of how Trump hurt the GOP candidates by firing in all directions instead of rallying the troops for victory in Georgia, you can’t get far toward the big goal if you stop to throw rocks at every stray dog along the way.

Until around a week before Election Day, the betting odds still heavily favored both Republicans, and remained remarkably stable. Their prospects also moved in tandem, reflecting their practice of campaigning together and hammering the common theme that their opponents were leftist radicals championing socialism. From Nov. 11 to two days after Christmas, Loeffler’s chances mostly toggled between 60% and 70%. As late as Dec. 22, Miller’s model put Loeffler’s odds of victory at 65-35, or almost two-to-one. Five days later, she still stood at a comfortable 60%. A free fall followed. By New Year’s Eve, her chances had slipped to below even at 48.0%, and by Jan. 2, she had dropped to just over 40%. In less than a week, Loeffler’s chances cratered by 20 points, taking her from a strong favorite to a moderate long shot.

Perdue followed a similar path, though his slide started later. His odds of winning were mainly over 70% from mid to late November, then stabilized at 60% to 65%. On Dec. 31 he was still a 60% favorite, and as late as Jan. 2, still ranked better than 50-50. By the eve of the election, his chances of winning had tumbled to around 45%. By Fortune’s estimates, his odds had fallen 15 points in just four days, a bit less than Loeffler’s, but foreshadowing disaster.

Why did bettors send both candidates’ odds swooning so late in the contest? One reason is that it was only after Christmas that the trend in early results came became apparent—and Georgians voted three ways. First, via absentee ballots sent by mail or deposited in drop boxes; second, by visiting polling sites starting on Nov. 14 in what’s called “advanced voting”; and third, by visiting the polls on Election Day. “The more data we saw, the more difficult people realized election night would be for Republicans,” says Bernard Fraga, a political science professor at Emory University. But that realization came slowly. At first, things looked good for the GOP. “In the first few days of early voting, it looked like white turnout was outpacing Black turnout by a higher margin than in the presidential election. The assumption was that the white share would gain over November, which would be typical in a runoff. If the Black share lagged versus November, that would be a bad sign for Democrats,” notes Fraga. That perception boosted Loeffler’s and Perdue’s pre-Christmas odds.

But in the weeks before Christmas, the trend reversed, and Black turnout claimed a higher share of the total than in the presidential race. “Then, the perception was that Republican voters would claw their way back after Christmas. It didn’t happen. Then after New Year’s. That didn’t happen either,” says Fraga. By that point, the Democrats had built a huge lead in the early voting. “The race on Nov. 3 was extremely close, but it became apparent that in the early voting, the Democrats were doing better than in the early voting in the presidential contest,” notes Chip Lake, a Georgia political strategist.

So a few days after Christmas, it dawned on the oddsmakers that the Republicans would need a super-huge turnout on Election Day to overcome the gigantic lead the Democrats had banked through early voting. Prior to Jan. 5, Georgians had cast 3.1 million votes that turned out to be 70% of the total. “That’s a mind-boggling number,” says Lake. Bettors quickly slashed Loeffler’s and Perdue’s odds of winning. Surprisingly, if the bettors had known that the Election Day turnout would greatly exceed the GOP’s best hopes, totaling 1.3 million, they might have kept backing the two Republicans. “If I’d known just before Election Day that 1.3 million voters would go to the polls, I’ve have predicted we’d win both seats,” says Lake. The rub: On Election Day, Black and other Democratic-leaning voters astounded the experts by flocking to the polls as well, blunting the big Republican better-than-expected showing.

Lake notes that since Republicans still won 63% of the Election Day votes, they might have won both seats had 100,000 more voters gone to the polls. That leaves the question political experts will ponder for years: Did President Trump’s strategy of quixotic attacks on allies in Congress and the integrity of our elections, rather than uniting the GOP behind the Georgia Republicans, suppress turnout and cost his party control of the Senate? His influence isn’t totally clear. On the one hand, Trump was raging for weeks that the presidential election was stolen without hurting the strong odds for both Loeffler and Perdue. The rallies he held for them in northern Georgia also appear to have contributed to their strong showings in and around the counties where the President appeared.

Fraga notes that any small change in turnout, especially given the closeness of the Ossoff-Perdue vote, could have saved that seat—and the Senate—for the GOP.

So on balance, Trump’s antics probably amounted to a negative by suppressing Republican turnout. “His attacks on the integrity of the system and how his demand for $2,000 checks for individuals backfired hit Republican turnout in the early voting period,” says Fraga. He adds that it wasn’t so much any one of Trump’s distractions, but the accumulation of the attacks that stoked warfare among Republicans and questioned whether folks’ votes really counted that likely quelled loyalists’ enthusiasm for voting. “The mystery is how high the Election Day turnout would have been if not for Trump’s rhetoric,” says Fraga. “It was high, but not high enough given the Democratic turnout. Maybe it would have been higher. It’s hard to know.”

It’s a good bet that Georgia isn’t nearly as blue as Ossoff’s slender and Warnock’s not-so-slender victories would indicate. Georgia appeared to be running true to form until just before the election, then both Republican incumbents’ odds collapsed in a couple of days. That shocking reversal could mean that U.S. policy for the next few years will be far different from what it would have been if the GOP had taken just one seat. Much of the blame probably resides with Donald J. Trump. The former President will likely pay the price for neglecting Georgia when the Democratic Congress begins to unwind his legacy.