All economic systems are to some degree adaptive and nonlinear, because people are adaptive and nonlinear. Economists generally create static linear models of these systems. Additionally, economists tend to assume Gaussian errors with diagonal covariance matrices because the math tends to blow up otherwise. Non-gaussian errors are fairly common unless your data set is a bunch of means (thus the ubiquity of "outliers"). Beyond that, economists assume "rational" agents that maximize utility, but this is a fairly poor model of human behavior.

Economists aren't uncovering an underlying truth, they're basically building toy models. These toy models might approximate economic systems in a narrow range of controlled circumstances, but things rapidly deteriorate outside that range. The problem is that people assume the models reflect reality, and make bets based on that assumption.

It would be nice if economists focused on fully Bayesian models with Gaussian process priors and long tailed (ideally Levy alpha-stable with a low alpha bias) likelihood functions. At least then the uncertainty of their models would be absolutely clear, and black swan type events would be partially factored in.