macrocephalopod

Well, look at your tweet thread: "Here's a quick primer on how to build a quant reversion strategy. First step is to find a price series that...may have some reverting behaviour -- this ... looks reasonable but you could use an ETF, currency, whatever really."...

Your tweet clearly implied that in order to make money there has to be some predictable behavior like mean reversion (or trend, presumably). But that's not true - a stock that wanders aimlessly with NOISE with a geomean multiplier of 1, so it goes nowhere - could make you rich.

I agree with that!* I have never disagreed with it! I didn't bring up shannon's demon -- I only pointed out that it doesn't work on price series that have zero arithmetic return (which is true of the price series I was using).

The * is there because this only works in theory (in reality transaction and financing costs would eat all your pnl) and only for certain definitions of "very rich" -- in practice you would need to compound for dozens of years to become rich, even if you had no txn costs.

Well, you're clearly making some assumptions on the (1+R) noise multiplier with geomean=1. If it's highly volatile noise, you could get rich as as fast as Rentech did with Medallion. And it's just random noise!

Yes I'm assuming it's similar to real-world stocks. If you took Bitcoin with 100% annualized vol you would only have a Sharpe of 0.5 so you need to compound for 16 years to be 95% sure of profit. For a stock with 50% vol you need to compound for 64 years to be 95% sure.

Shannon's original example used 50% vol for each period -- if we assume that a period is one day, the asset in his example would be 8x more volatile than Bitcoin (!)

I honestly don't know what the point of these arguments is. Yes, there are some theoretical models where you can make money arbitrarily fast if you assume completely fantastical parameters and ignore all real-world frictions. How is this supposed to be helpful?