I often talk here about decisions I/Alameda made that went well. Sometimes people ask for examples of the opposite -- times when I made a mistake and lost a lot, or even times I lost a lot by doing the right thing. Both happen a lot!

A thread about melted wings.

Alameda uses "March 12" in a Voldemort-like way -- it invokes dread like not much else, and it comes up a lot in mean vs. median discussions ("sure this usually works but it loses $5m on March 12," etc.)

Before March 12, though, there was another Terrible Day: September 25.

September 25, 2019 was -- at the time -- the scariest day I'd ever had trading, and I think it was maybe Alameda's worst potential vs. realized PNL day ever (we *could* have made a TON -- we, uh, didn't).

Reading through Slack from that day, there's sorta weirdly little written down -- everything, it turns out, was getting yelled frantically on the trading floor. I did screenshot one highlight though :P

Anyway: what happened that day?

Really, it was classic. There was some organic selling, and then there were a TON of liquidations, and that drove the market down a LOT. Because the liquidation-driven selling was inorganic, a lot of the resultant downturn reversed very fast.

The biggest source of liquidations back then was BitMEX BTC perps -- I used to spend a lot of time just sorta staring at that order book, moving our deltas in response to big changes -- to good effect!

A 20% drop impacted the whole market, though -- liquidations happened everywhere, and various futures got to GIANT discounts as a result. Anyone with a ton of free capital could make a KILLING trading these -- buying the cheap futures vs. selling spot was even delta-neutral!

So, what went wrong for us? I've mentioned we used to maintain delta-neutrality. Our capital base (at the time) was heavily concentrated in BTC. To get, for instance, spot USDT to trade with, we needed to sell some of that BTC, making us naturally short all the time.

As a hedge, then, we were always long something else on margin -- often, for instance, BTC perps or futures. And back then, we sometimes played things a little fast and loose, collateral-wise.

What was sort of going on: we can always lever up and put on bigger positions than we are, the trade-off being higher probability of liquidation. So, maybe we're choosing between size X and 30% distance to liquidation vs. size 2X and 15% distance to liquidation.

Most days? We're gonna make twice as much by doubling all our sizes. But on the rare days when the market *does* move 15%+? We're gonna have to mobilize additional collateral FAST, or we're gonna get fucked.

15% was a number we let ourselves reach when trading was *really* good -- and, in the days leading up to the crash, trading had been *really* good, with futures premia moving all over the place. But then the market crashed 20% which is > 15%, so we were in hot water.

We'd morphed into an Icarus.
We had flown too close too the sun.



(couldn't help myself here)

The crash was *quick*, so, while we have systems in place for getting more collateral to the places we need it most, we couldn't save everything. Normally we'd try *really* hard to buy when, e.g., OKEx quarterlies get to a 10% discount (they did on September 25) ...

... but this time, at least at the beginning, we sold some via liquidation :P (I don't remember the exact products or numbers here, but it was something like this)

So while we were doubling our $ on most days, we were kneecapping ourselves on the most important days of all.

This was a crucial lesson for us! We hadn't really had to think hard about how valuable these giant move days are -- now we know how much we could make when they happen (200x the PNL of a then-normal day, conservatively).

That means if a day like that happens > 1/200 of the time (which seemed likely, and is true), it's immediately good to optimize for those days instead. And that's disregarding that you might lose a lot to getting liquidated -- which, on the positions where it happened, we did :P

And that meant halving our typical position sizes, and keeping a TON more spare capital around to put on giant futures positions when they got to these 10% discounts ...

... like they did again on March 12! And occasionally have for short periods in the time since.

We had NO capacity to do the best trades on September 25, and that's the thing that hurt way more than getting liquidated. We'd gotten greedy for not enough edge which meant we couldn't maximize when it mattered most -- and we paid the price.

But from lemons we made plenty of lemonade. We changed our strategies a lot in the days/years following and have been able to do a lot better during high volatility ever since -- decreasing our median PNL but increasing our mean by a ton. And the E in EV only cares about means.

Mistakes are natural in trading -- especially crypto trading where everything is so fucking weird. It's what you learn from your biggest mistakes that matters most, though -- and we've (thankfully) learned a lot.