01 May 2017

Fruit Pies

Let's do some dot connecting.

1 - United Airlines has some rent-a-cops manhandle a elderly doctor out of his seat and off the plane. United eventually settles for an, as yet, undisclosed sum.

2 - Southwest says it will cease overbooking. The same report also says that, at 15,000, Southwest had more passenger bumps than any other airline.

3 - Airline reservation systems, computer division, go back to the 1950's, but the most well known, ACP (some might argue SABRE, but that's just a specific implementation), goes back to the mid-1960s. Of note, it continues today, in much modified form, as TPF.

The distinguishing feature of both ACP and TPF is that they're, at heart, assembler programs with bespoke "transactions" embedded in OS files. Very lean, very fast. According to reports from 2001, SABRE was going to port to Compaq machines.

Much has been made here about the many failures of quant methods when dealing with human systems/processes. When doing data from Mother Nature, who doesn't change the rules every now and again, quant works rather well. In the human realm we get The Great Recession.

So, the dot connecting exercise leads one to conclude that the quants doing computer reservation haven't paid much attention for the better part of 60 years. It's an exercise one might (well, can, actually) find in an upper-level undergraduate program (11.128): on a per-flight basis, calculate the percent overbooked such that only 5% or 1% or 10% of flights actually bump passengers. What might be called low hanging fruit.

The airlines have all the data needed to accomplish the exercise. They know the actual load for each flight going back forever. They know the external factors which affect variability of load: holidays, weather, conventions, and so on. So, why the United and Southwest experiences? The only logical answer is simple: airlines have found that they can get away with not paying attention. Where's a competent data scientist when we need one?

No comments: