This is a very vague question. I’m curious how pricing is done in industry, specifically at large tech companies. I know that Amazon does not price discriminate based on user, so price experiments are executed over time.
But in a broader sense, I’m wondering how the willingness to pay distribution is fit. If you only ever charge one price, you don’t have enough data to estimate the curve.
More broadly, if you assume a linear model you get positive demand for negative prices and negative demand for arbitrarily high prices. Using a log transform can help, but now there is issue of heteroskedasticity in dollar (correlation between dollar amount an error.) and a logistic model assumes a fixed market size (40% of the market will demand product at this price) but assumes that the market size can be estimated well externally.
Additionally, with demand forecasting, this gets even more complicated as the market size fluctuates (less people in market for pool supplies in winter.)
Given all the thorny problems here, are parametric models preferred? Or more flexible ML approaches?
What have you seen gain traction in industry?
