Better thinking through technology

An autopsy of the 2014-15 NBA bot



I'm pulling the plug on the NBA model for the 2014-15 season, as the two games a night that we now have aren't presenting very viable opportunities to do anything except piss away money. As we approach the finish line, there are signs that the model is breaking down as teams that are losing keep fouling to the finish and running up their opponents' spreads. It looks like the Nets upsets of the Hawks were the last good bets of the year, and I doubt there will be many more predictable upsets this season that are worth pursuing.

It should be noted that major changes were implemented about six weeks ago, and I'll discuss those separately, as they significantly improved the quality of the bot's performance.

That all said, let's look at the overall numbers first. These numbers only cover pre-match bets that were placed and publish before each game. This does not include any numbers for live betting, as I feel it would be unethical to discuss data that wasn't published prior to the games that were played.



Overall Numbers




Total run time: 14 weeks

Total bets: 574

Total correct bets: 261

% correct: 45.47%

Net profit: 1.92% (bear in mind I was betting lines that went as high as +1700, so a 50-percent hit rate is not needed to turn a profit)



Modified Model Numbers




Total run time: 5 weeks

Total bets: 147

Total correct bets: 77

% correct: 52.38%

Net profit: 9.85%



Investment Worthiness Analysis




First, let me lay down what I use as my default way of valuing anything as an investment. I measure investments against a single stock, and that stock is Altria. (Full disclosure, because I am mentioning a security, I have maintained a long position in Altria for several years.)

Why Altria?

Altria is a company that has historically gone up in value and boosted its dividends. For the sake of analysis, let's assume that we're only looking at the dividend. Altria was trading at $50.44 at last check today, and that comes out to a dividend yield of 4.1%, which is widely considered a high dividend, especially for a stock that has doubled in the last couple years.

So, that 4.1% rate of return makes an excellent baseline for what one might consider relatively worry-free investing, or as close as one can realistically assume with any at-risk investing position.

So . . . what does this tell us about the NBA model as an investment? The answer is mixed. The initial model was clearly below investment-grade, tracking between break-even and a 2-percent profit. The second model, while not thoroughly tested, shows signs of being worth about twice the value of a solid investment-grade, boring stock like Altria.

In short, the conclusion regarding NBA betting is that the data is inconclusive at this time, but encouraging enough to merit another year of betting.

For reference, this is better than the NCAA men's basketball model, which got very ugly toward the end, turning from a small profit to a small loss. The NBA model is more consistent, and that's likely due to the high quality of play within the league.

As added reference, however, even the second version of the NBA model returned slightly more than half the profit that the NCAA football model generated. The automated NCAA men's model, which only targets ML bets, produced a total profit of 16.73% across 12 weeks.

Overall, the best that can be said for the NBA model at this time is that it produces a tolerable enough return to bridge the gap between football and baseball seasons, and it may yet have potential to produce better than investment-grade results next season.

I am very curious to see how well the model performs in WNBA play, where one would expect the betting public to suffer from a certain level of informational poverty. I will, however, note those hopes may be misplaced, as basketball is one of the most widely and accurately model sports in all of betting.

I am also curious to see if the second version NBA model can be converted into something useful for betting college basketball games.