When it comes to catastrophe loss estimates from events like Hurricanes Harvey and Irma, odds are insurers and others will have to wait quite a while before the true cost of the event is known.

Catastrophe models generated a wide range of estimates of insured damage in the immediate aftermath of Harvey, all of which are sure to be revised and revised again. While the models often differed, all indicated Harvey’s toll would fall considerably short of those caused by the three costliest hurricanes in recent U.S. history.

The Insurance Information Institute, citing research conducted by Property Claim Services, a Verisk Analytics business, puts 2005’s Hurricane Katrina at the top of the list, with $49.79 billion in insured damage in 2016 dollars. The next costliest, 1992’s Hurricane Andrew, caused less than half as much, with $24.48 billion in insured damage in 2016 dollars, while 2012’s Superstorm Sandy came in at $19.86 billion.

The highest initial insured-damage estimate for Hurricane Harvey came from Karen Clark & Co., which put total insured loss—not including losses covered by the National Flood Insurance Program—at roughly $15 billion. This included about $2.5 billion in wind damage, $500 million in storm surge and more than $12 billion in inland flooding in Texas and Louisiana.

Modeler AIR Worldwide estimated insured losses—again excluding properties covered by NFIP—would exceed $10 billion, with $3 billion stemming from wind and storm surge. But the insured losses amount to only a fraction of total property losses, which AIR estimates could hit the $65 billion to $75 billion range.

CoreLogic, another modeler, estimated private insurers would sustain less than $500 million in insured flood losses, while wind losses would range in the $1 billion to $2 billion range. Uninsured flood loss could reach $27 billion.

Damage to private and commercial automobiles as a result of Harvey is expected to be relatively high. A spokesman for the Insurance Council of Texas said insurers faced an estimated $2 billion in losses from private automobiles and an additional $1 billion from commercial auto, including mobile homes and trailers.

One critical question is how big a loss the nation’s largest flood insurer—the federal government’s NFIP—will ultimately take. The program was already more than $24 billion in debt when Harvey struck. The program enjoys—if that is the right word given the potential losses—a virtual monopoly on residential property flood insurance and underwrites commercial properties as well. Yet both CoreLogic and modeler Risk Management Services provide remarkably similar answers, with CoreLogic estimating losses of $6 billion to $7 billion and RMS estimating $7 billion to $10 billion. As of mid-September, NFIP had not released its own estimate.

RBC Capital Markets issued an analysis that looks at the historical pattern of loss estimates from catastrophes in general since the terrorist attack on 9/11. “They start a little too low, then they seesaw to being a little too high and often end up towards the higher end of the gap in between,” RBC says.

The company uses as an example the course 9/11 loss estimates took. Initial loss estimates were in the $10 billion to $20 billion range, but “conventional wisdom quickly adopted $50 billion as the going rate.” The final tally was closer to $30 billion. Estimates for Hurricane Katrina followed a similar pattern.

“In our experience, this is the ‘normal’ reaction—to favor the high end of the range until information crystallizes,” says RBC. “As one client commented, ‘taking the over’ on loss estimates has always been the safe play.”