In this episode of "How I Wrote This," Dan McCarthy and @Wharton's Peter Fader break down their groundbreaking JMR paper on "Customer-Based Corporate Valuation," a strategy that's already getting buzz from @HarvardBiz and Wall Street.
Listen on Spotify:
https://t.co/0ID84qkyQG
Hold on! Before we give up on gamma distributions, etc., let's also consider model parsimony, robustness, comparability, and inference. The scenario described below will lead to horrendous overfitting and all kinds of associated problems. Occam's Razor still applies!
niche post but so much of undergrad probability / stochastic processes is working with distributions that subpar representations of a certain model but have quite nice analytical properties (eg Gamma)
I suspect this requirement is gone with Cursor and we will focus on purely computational models that maximize fit to reality versus analytical tractability. This has probably been the norm among engineers for many years (? I suspect but am not sure) but will now become the norm across all mathematical disciplines thanks to Cursor
like "Okay just sample from Lognormal 75% of the time and Pareto(a=3.5) 25% of the time and let me see what that looks like"
This may be completely obvious to engineers but is a huge upgrade for those of us who studied economics / any modeling discipline pre-2020 and had to solve these things by hand and were extremely limited in the models we could build as a consequence
niche post but so much of undergrad probability / stochastic processes is working with distributions that subpar representations of a certain model but have quite nice analytical properties (eg Gamma)
I suspect this requirement is gone with Cursor and we will focus on purely computational models that maximize fit to reality versus analytical tractability. This has probably been the norm among engineers for many years (? I suspect but am not sure) but will now become the norm across all mathematical disciplines thanks to Cursor
like "Okay just sample from Lognormal 75% of the time and Pareto(a=3.5) 25% of the time and let me see what that looks like"
This may be completely obvious to engineers but is a huge upgrade for those of us who studied economics / any modeling discipline pre-2020 and had to solve these things by hand and were extremely limited in the models we could build as a consequence
@TheiaResearch Hold on! Before we give up on gamma distributions, etc., let's also consider model parsimony, robustness, comparability, and inference. The scenario described above will lead to horrendous overfitting and all kinds of associated problems. Occam's Razor still applies!
▶️ Watch now: https://t.co/e67fqZqUmN
@Wharton Professor @faderp, in Customer Centricity, argues that “the customer is always right” is outdated. Even giants like @Starbucks grow sustainably by focusing on the right customers, not all customers.