Alameda won't be participating in this, but it does present a chance to explain how we think about the value of man vs. machine in our trading. https://twitter.com/FTX_Official/status/1265128296217370625

People hear that Alameda’s trading is “quantitative” and often assume that means nothing we do is manual, and all of our trading centers around automation. While we fairly rarely e.g. use an exchange’s website to place orders, we do a lot I’d call manual, too.

On the machines’ side, the vast majority of our trades are done by automated strategies we call bots. A lot of the work we do is in optimizing those bots -- finding and optimizing data sources, implementing and refining trading strategies, etc.

We’re constantly iterating by running studies to identify issues and opportunities, back-testing strategies, auditing new and existing data sources, and thinking about innovative trading strategies to stay ahead of the competition.

We also spend a lot of time optimizing specifically for robustness -- if our bots work perfectly 99% of the time and fail unpredictably 1% of the time, trying to trade as much as we do can expose us to a lot of downside, so we have to avoid it.

And, at the end of the day, doing a lot of market making and trading every major product on every major exchange is always going to require your set up to be largely algorithmic.

But while machines underpin most of our trading, man is still a key piece. Every day, our trading team makes hundreds of on-the-fly decisions which we consider essential, and without which it’s clear we could never maintain the success we have.

(Also, while I’m saying “man” for rhetorical reasons, I should note that Alameda’s trading team bucks industry trends by being something like 20% women, which is part of what makes it by far my favorite trading environment I’ve ever worked in.)

Tons of parameters feed into our models, and we change many of those manually really often -- we control things like position sizing, risk, interest rates, and various volatility parameters mostly manually in response to changing market conditions.

How long or short to be, for instance, is mostly determined manually -- sometimes we’ll want to hedge something specific, or have some idiosyncratic reason we believe we have an edge in making directional bets (understanding funding or options expiration flow, e.g.).

Additionally, a ton of our capital management is done manually -- from determining the most profitable opportunities and sending capital there to take advantage, to analyzing where we need to send capital to mitigate risk, to figuring out where we need money to settle OTC trades.

All of our “truffle sniffing” is also pretty much just manual -- stuff like rooting through exchange margin minutiae, searching for rare super valuable arbs, checking whether every last dumb-sounding opportunity is actually great. This can feel thankless, but often turns up gold.

And this is just on “typical days” -- during the most insane 1% of trading, we know our models are trained much more for median behavior and we don’t expect them to perform perfectly.

So we override lots of things with super aggressive parameters by hand when we KNOW we understand e.g. BitMEX liquidation behavior better than both the market and our bots, and positioning correctly can be the difference between making or losing 5% on a huge position.

These days are why I LOVE crypto trading -- someone’s always screaming what we need to do (“BUY ANYWHERE WE CAN!”, “SEND ALL OUR FREE BTC TO HUOBI!”, “BORROW AS MUCH F%^%ING USDT AS WE CAN!!!”) and everyone feels urgent -- because everyone is.

Anyway, this was really high-level and I’ll try to talk about more specifics re: all these topics some time. Just wanted to give a broad overview of what Alameda’s trading is like -- and how it’s the marriage of human thinking and computing power that gives us our edge.