Thread by @therobotjames

Playing the easy games at the right time at scale. 🧵

In the early 2000s internet poker was just getting started and a vast number of the players were terrible and, at certain times, likely to be drunk.

You could do very well by simply "not doing dumb stuff".

Thread 👇👇

1/n

Some of the best games were "sit and go tournaments".

These were short single table 10 handed no limit hold'em games, with rapidly increasing blinds.

If you win you got half the prize pool, second 30%, third 20%. Everyone else got nothing.

Something like that anyway.

2/n

Optimal strategy against terrible drunk players is pretty simple.

You play extremely tight to start: only playing when you are prepared to bet your whole stack. (People would call you with anything.)

And when blinds get big you get more aggro, isolate players and steal

3/n

If you understood who you are playing (on average) and are prepared to accept outcome variance, it was easy to make money.

Even if you were a very mediocre poker player, as I was. And still am.

https://twitter.com/therobotjames/status/1419065414768676869?s=20

4/n

Now you have an edge, you think about scaling.

You want to play as many of these games as you possibly can.

Your edge is real - but the outcomes are very variable.

So you want to play as much as you can to harness the edge as smoothly as possible.

5/n

But there's only 1 of you.

And you can only play so many games at a time.

And playing tight poker against drunk people is boring.

So you think - could I automate this so I can scale it effectively?

6/n

Now, humility is required.

You probably do some subtle things in your play you think add edge.

But a bot isn't gonna be able to do subtle things.

You need to work out what is essential to harness the edge - and do that systematically in the simplest way possible.

7/n

So back in 2002, wearing a plaid shirt and listening to Nickelback's "How You Remind Me" no doubt, I built a bot to play these tournaments at scale.

It had two moves:
- Fold / Call for free
- Bet entire stack

People HATED it! Watching people get mad in the chat was fun.

8/n

It was super dumb.

It would sit the first rounds out unless it got A-A or K-K in which case it would go all-in.

When blinds were big it would full-aggro steal blinds and bet good hands.

It was a very bad poker player.

But it didn't need to be better to beat these games.

9/n

And it could scale, playing loads of small buy-in games at the same time.

It looked super dumb.

But it was the appropriate, pragmatic way to harness the massive opportunity edge that existed in that market at that time.

10/n

'Playing dumb "good enough" poker at massive scale'

was a better move than

'Playing better poker at a smaller scale'.

Trading can look like this too.

11/n

If you have a good edge, you don't need to harness it in the most sophisticated way possible.

It can be better to do "roughly the right thing at roughly the right time, at scale."

This is why, like my poker bot, quant trading sometimes *looks* super dumb and blunt.

12/12