As I've been saying a lot lately on Twitter, Alameda is quite comfortable with the Tether team, and we do a lot of large creations! I talked about that in more length in this thread: https://twitter.com/AlamedaTrabucco/status/1348773027639562241?s=20

What does "large" mean, though? Well, it depends! And it doesn't just depends on any absolute metrics, or on market conditions, or whatever -- it also depends on our own ability to think through the question of "how much should we create" and "how much is like, too much, c'mon."

Soon after we got set up to do creations, we did some initial studies to determine how much we could sell above $1 + X + Y + Z, where X was creation cost, Y was "execution cost," and Z was "cost of tying up capital." We also charge for our time and add some edge somewhere.

X: this fee has varied over time a bit, but ultimately is pretty easy to understand and stays pretty constant almost all the time. FWIW, X != 0 is why the USDT premium can persist, since there's technically no way for people to just transact with USDT for $1 infinitely.

Y: this is where size starts to matter! The legs of this trade are: sell something (probably BTC) into USD, create USDT with USD, use the USD to generate something (probably BTC), and end up back where you started. And if the trade's still good, cycle! (We often do this).

The more size you're doing, the more impact each of these legs can have (selling more spot BTC into USD, buying more spot BTC with USDT) -- and that increases the cost of execution, Y. We've run studies to determine this cost Y for various sizes.

Z: size also matters here, actually! At any point we have some amount of capital floating around, free to do "the best trades" or just to chill if we think the best trade is "wait for something amazing to pop up and have capital ready for it."

Starting the USDT cycle requires a capital investment -- sending to create, and then that capital is being used in the various legs for some period. The first $1 we lock is less costly than the 50000000th since, if size > our free capital, we need to unwind positions, e.g.

And we've run studies about this, too -- but mostly this is a human call, since the value of spare capital varies so much with market conditions (when things are volatile, keeping spare $ around to do awesome trades with is better, for instance, than when it's slow).

So! This gets us to a point where we understand how much a given size will cost us (in fees, execution, opportunity cost). We compare to the premium we expect to sell the USDT at, and if it's good we do it! And we do the size S maximizing S * average profit.

Originally, though, we were getting this *quite* wrong -- we were being too conservative in estimating average premium, and we were over-estimating capital cost because we thought it would take longer to cycle. So we were under-creating by a lot.

Having gotten a ton of experience actually cycling, though, gave us WAY better data about premium impact and execution cost -- and better intuition about timing and opportunity cost (having the $ in USDT also isn't, like, worth 0 in a big move, which we were getting wrong).

We're better at this now -- and we're maximizing our bets a lot more efficiently when the premium and market conditions line up the best.