Best Twitter threads of the day – October 12th

I tracked several of the Celsius Wallets which stole MILLIONS 💰from investors, right before bankruptcy, including Alex Mashinsky

I'm making them all public in the hopes that the bankruptcy courts claw back these funds. Regular Investors > Executives. 🧵

— Coffeezilla (@coffeebreak_YT) October 11, 2022

Alex Mashinsky cashout 2 (USDC edition)
He stole 1,792,255.766134 USDC on 05/15/22. Once again it goes to a Coinbase wallet. Once again it comes from multiple wallets funded by CELSIUS NETWORK: WALLET 5.
Whether this is normal routing or an attempt to obfuscate funds is unknown. pic.twitter.com/2s8Sg2jdjV

— Coffeezilla (@coffeebreak_YT) October 11, 2022

Next up another Mashinsky related one: Koala 1 LLC who took $1,638,303.3376 USDC on 05/27/22. yet again funded from wallets funded by Cel wallet 5.
This one also went to Coinbase (Mashinsky seems to love @brian_armstrong for some reason), and once again the amount matches up. pic.twitter.com/CHgLaPIh8R

— Coffeezilla (@coffeebreak_YT) October 11, 2022

Now lets move onto other characters: Aliza Landes who is associated with Leon Daniel.
She withdrew $333,547.85244 USDC on 05/29/22
Doesn't take long to find her withdrawing to Kraken coming from a wallet funded by CEL wallet 5. pic.twitter.com/1DbQh6pvnQ

— Coffeezilla (@coffeebreak_YT) October 11, 2022

Leon Daniel withdrew $2,370,633.82 USDC on 05/29/22 going to Kraken… coming from Cel Wallet 5.
I can't believe these people thought they'd get away with this. pic.twitter.com/w7J8UQdo9X

— Coffeezilla (@coffeebreak_YT) October 11, 2022

Summary: LOCK THEM UP
Alex Mashinsky Eth
0x3001ccC98B917c0f604E90C94B4E2Ed8a93c2281
Alex Mashinsky USDC
0x33BBf8B3885214c0CE53698D77Ea0d292906c0Bb
Koala 1 USDC
0xBC12B94EC3B7AAFF9d91428DfCB505c7c553602F
Leon Daniel USDC
0x884fF7A885F7EA52230670FcAE5e05fa24B68b2c

— Coffeezilla (@coffeebreak_YT) October 11, 2022

Aliza Landes USDC
0x28183C9BD76AE0d87D36Fd2955a58B2f70968Dc2

— Coffeezilla (@coffeebreak_YT) October 11, 2022

1) When it comes to oracles,

you just have to make up your own damn mind pic.twitter.com/7kZATSLpQM

— SBF (@SBF_FTX) October 12, 2022

2) NOT FINANCIAL ADVICE, NOT LEGAL ADVICE

— SBF (@SBF_FTX) October 12, 2022

3) Let's say that you have a margin trading protocol which allows cross-margin between a bunch of different assets.

Your algorithm is "require 30% initial margin and 20% maintenance margin"–i.e. allow ~3x leverage, and margin call accounts if they get up to 5x leverage.

— SBF (@SBF_FTX) October 12, 2022

4) You have an 'oracle' which reports the price of the assets, and your risk engine–the thing that calculates users' margin–uses prices from that oracle.

One asset–say XYZ–has a price of roughly $0.40 according to the oracle.

A user–Alice–has $200m XYZ and -$25m USD.

— SBF (@SBF_FTX) October 12, 2022

5) Alice:

500m XYZ ~ $200m
-25m USD = -$25m
net value ~ $175m
margin ~ 87.5%

how much should the risk engine allow Alice to withdraw?

— SBF (@SBF_FTX) October 12, 2022

6) A naïve answer would be "well if you require 30% margin then Alice can withdraw up to $115m; at that point her account would be:

500m XYZ ~ $200m
-140m USD = -$140m
net value ~ $60m
margin ~ 30%"

— SBF (@SBF_FTX) October 12, 2022

7) But there's a picture of a mango over The Oracle's face.

So, obviously, XYZ is actually MNGO, and MNGO isn't really worth $0.40; it's worth around $0.04.

Alice's real account value is, roughly:

500m XYZ ~ $20m
-25m USD = -$25m
net value ~ -$5m
margin < 0

It's already gone.

— SBF (@SBF_FTX) October 12, 2022

8) But for a brief period, someone (almost certainly Alice) bought enough MNGO that it did trade for $0.40, and before people could get in to sell and return it to a more reasonable price, the risk engine had already allowed Alice to withdraw $100m from Mango Markets.

— SBF (@SBF_FTX) October 12, 2022

9) So, what went wrong? Did the oracle fuck up?

I mean, not really. Or, rather, it depends on what the specs of the oracle were.

The oracle accurate reported the current price of MNGO. It's just that the 'current price' wasn't really anything close to the 'fair price'.

— SBF (@SBF_FTX) October 12, 2022

10) What should you really do?

Well, let's look at what FTX does, and why *it* didn't have any risk engine issues.

— SBF (@SBF_FTX) October 12, 2022

11) First, FTX has EWMA price bands.

What this means is, roughly, that FTX consumes raw price feeds.

But, before feeding that into its risk engine, it bounds those price feeds so that they can't move more than ~20% over a 5 minute period.https://t.co/Owu4xyesPB

— SBF (@SBF_FTX) October 12, 2022

12) The reason, basically, is:

b) If an asset's price moves 3x in 1 minute, it's decently likely to be bad data, or a temporary wick, or something like that

b) If the asset's price stays there for 5 hours, that new price is likely the true economic price

— SBF (@SBF_FTX) October 12, 2022

13) FTX's MNGO index price moved–there should be *some* update–but much less than others because of those EWMA price bands.

In particular, while FTX's index topped out at a ~100% increase, on some exchanges the move (temporarily!) hit a +900% increase (!!!). pic.twitter.com/ZrfpjveL7w

— SBF (@SBF_FTX) October 12, 2022

14) Second, FTX uses 'IMF Factors'.

The larger your position, the greater % margin we charge.

For MNGO, the margin we charge is 0.00025 * sqrt(MNGO tokens).

If you wanted to have a 500m MNGO position, FTX would have required, uh, 500%. (Bounded at 'fully funded'.)

— SBF (@SBF_FTX) October 12, 2022

15) The reason, basically, is that large positions–especially in illiquid tokens–can have a lot of impact.

So we charge more % margin the greater your position is.https://t.co/d14JDRnHYd

— SBF (@SBF_FTX) October 12, 2022

16) And some positions — like the one in question — are large and illiquid enough that the risk engine forces you to fully collateralize a position.

— SBF (@SBF_FTX) October 12, 2022

17) So even before hitting position limits, the risk engine ensures that the collateral backing a position is sufficient.

And what if you try to use something other than dollars as collateral?

Well, we haircut it. In some cases, a lot.https://t.co/yeML2zCkgg

— SBF (@SBF_FTX) October 12, 2022

18) (It's worth noting that this is all referring to FTX International.

In addition to all of the above protections, our amendment for FTX US Derivative's margin order would only be for BTC and ETH futures using USD margin. No MNGO, and certainly no MNGO collateral.)

— SBF (@SBF_FTX) October 12, 2022

19) There are a bunch of other risk engine protections and sanity checks, too, which would have caught something like this.

So — back to the oracle.

It reports:

"MNGO: $0.40"

Is it wrong?

— SBF (@SBF_FTX) October 12, 2022

20) Well, it depends on what it's promising.

But probably it's just promising to tell you, literally, what MNGO is currently trading at. And, for a brief period, on some exchanges, MNGO was in fact trading at $0.40.

The real problem here was using the raw oracle price.

— SBF (@SBF_FTX) October 12, 2022

21) The Oracle tells you everything and nothing–the history and current state of markets.

It's the risk engine's job to consume that information, and decide what positions are safe.

Sometimes it can't just regurgitate The Oracle. Sometimes it has to make up its own damn mind.

— SBF (@SBF_FTX) October 12, 2022

22) Which doesn't mean that the risk engine needs to be manual.

You can create a set of rules for it so that it's conservative, and handles apparent large moves gracefully.

That, in the end, is probably the most important thing we do at FTX.

— SBF (@SBF_FTX) October 12, 2022

23) And, really, it's why we started FTX in the first place.

Tradfi had sophisticated (sometimes!), slow, manual risk models, and–in some FCMs–fast, egregious ones.

Crypto had fast, automated, broken risk models.

There was an opening for a thoughtfully automated risk engine.

— SBF (@SBF_FTX) October 12, 2022