While thinking about economic simulations, I couldn’t help but notice their similarities to climate prediction simulations and weather forecasting. They are sensitive to the data used to define their initial situations and they can produce different outcomes even when using the identical initial setups. This variability of outcomes is required because the real world is NOT exactly deterministic. Sometimes the panther jumps on the baby antelope and the antelope escapes anyway, despite the odds being against that outcome. Sometimes someone is pulled out alive from the rubble two weeks after an earthquake. Sometimes it doesn’t rain on the parade, even though it should have.
This variability is generally accomplished using random numbers that fit some range of expected values. But randomness isn’t enough. The variability should also have a probability associated with the value selected for the range of expected values. Randomness assumes that if you dropped a marble on a plate, the atoms in the marble and in the plate could randomly line up at some point and the marble would fall right thru the plate. Probability says that you’d have to drop the marble a lot of times before their atoms would align well enough that the marble would fall thru the plate.
So any economics simulation should be able to meet these twin requirements of variability and probability. The starting numbers should affect but not determine the final outcome. The final outcomes should be different for the various runs but their outcomes should fit into some range of final outcomes that match a probability curve that makes sense.
For example, an individual might start off in poverty and (rarely) grow to great wealth, but if the simulation shows it happening most of the time, there’s something inherently wrong with the simulation, something that produces an unrealistic outcome. It doesn’t mean that there can’t be circumstances and coincidences that allow for happy outcomes, just that they shouldn’t happen most of the time.
Just like every time a cloud approaches it doesn’t mean that a flood is going to wash out the inhabitants of a valley, valid and useful economic simulations have to be able to produce a range of possible outcomes approximately in proportion to their real world probabilities.
Once we have an economic simulation capable of producing a stable range of probable results, we will be able to change the operating parameters (raise minimum wage or reduce taxes) and study the range of results that are produced. If they produce unexpected results, either our software is wrong or our expectations are wrong.
My suspicion is that it wouldn’t take too long to wring out the errors in the simulation software because of the many sets of eyes with great vested interests in the results. And once the simulation is cleaned up, economics will finally be able to move away from the Farmer’s Almanac version of weather prediction and into a time where the effects of economic policy changes can be predicted with the same level of precision as weather is predicted now.
Consider how much weather prediction has improved in the past 50 years. Weather predictions used to be, “It will probably rain next week” to “we will probably see some light rain starting late next Tuesday and get heavier overnight into Wednesday and will probably be gone by Thursday morning.” Sure, there’s always some variability, such as when the rain starts Tuesday noon instead of late afternoon, or doesn’t start until Wednesday morning, but for the most part they are much more accurate than they ever used to be.
I look forward to when we can expect the same improved accuracy in our predictions of the impacts of changes to economic policy.