Demand surge is understood to be a socio-economic phenomenon of large-scale natural disasters, most commonly explained by higher repair costs (after a large- versus small-scale disaster) resulting from higher material prices and labor wages. This study tests this explanation by developing quantitative models for the cost change of sets, or "baskets," of repairs to damage caused by Atlantic hurricanes making landfall on the mainland United States. We define six such baskets, representing the total repair cost, and material and labor components, each for a typical residential or commercial property. We collect cost data from the leading provider of these data to insurance claims adjusters in the United States, and we calculate the cost changes from July to January for nine Atlantic hurricane seasons at fifty-two cities on the Atlantic and Gulf Coasts. The data show that: changes in labor costs drive the changes in total repair costs; cost changes can vary significantly by geographic region and year; and cost changes for the residential basket of repairs are more volatile than the cost changes for the commercial basket. We then propose a series of multilevel regression models to predict the cost changes by considering several combinations of the following explanatory variables: the largest gradient wind speed at a city in a hurricane season; the number of tropical storms in a hurricane season whose center passes within 200 km of a city; and cost changes in the first two quarters of the year. We also allow the coefficients of the regression model to be stochastic, varying across groups defined by region of the Southeastern United States and year. Our best models predict that, for any city on the Gulf or Atlantic Coasts in any hurricane season, the residential total repair cost changes vary from 0.01 to 0.25, depending on the wind speed and number of storms, with an uncertainty of 0.1 (two standard errors of prediction) given the wind speed and number of storms. The commercial total repair cost changes vary from 0.005 to 0.15 with an uncertainty of 0.08. Our models including wind speed, the number of storms affecting a city, and cost changes in the first half of the year explain roughly half of the observed variability in cost changes. Additional explanatory variables that we have not considered may account for the remaining variability. Given these models, however, there is still considerable uncertainty in their predictions. This uncertainty arises from variations between groups defined by region and year, not from variations within a given region and year.