Ever wondered how the pros do CEX-DEX arbitrage? Here's a deep dive on a $2.5 M bot's strategy and execution.
Read the full breakdown👉 https://t.co/f13gyUNNqt
A lot of people ask me how to build good trading systems. I usually recommend this book, not for designing your own system but to understand how exchanges design theirs and where the flaws might be.
Read this, then try to figure out where those systems fail.
https://t.co/4ad7Go9adU
.@Uniswap settled 140 million trades in August, more than Cboe BZX and NYSE American combined.
NYSE American has been around since 1908. Uniswap launched in 2018.
Five years ago it was ~2 million trades a month.
Next rung up: NYSE Arca, at 232 million.
Here’s a fun one I found on Robinhoodchain.
If gas limit is above 4m it shows fee of 0, but if lower it charges a random fee
Aggregators simulate with very high gas limits but wallets execute closer to expected.
It also has a additional setter to force fees on certain addresses that is used to force arb bots that have figured out how it works to still have to pay
Update:
Those agents are super pessimistic. The main problem is they skip actual alpha, claiming it is “not tradable.”
Best win so far is to push them to do whatever it takes to realize an opportunity.
At the current pace the agents should pay for themselves this month, maybe break even including server costs.
🧵[1]
A slightly weird experiment I’m running:
AI agents have already found usable MEV alpha for us with me guiding the research.
Now I want to see how much of me is actually necessary.
I’m giving one our research history + code, $100 for gas/compute and a VPS.
@s1csss The best tip I can give you is to freeze the chain a specific arbitrage opportunity and brute force many profitable cycles.
Then look at the data and try to understand what are the features you’re looking for (liq, ticks, fees, amount in/out curves etc)
We are opensourcing the way we simulate transactions on @opcode_fi. It's an extension on top of reth, that simply streams all ethereum state changes over grpc. Any cold state is loaded through rpc as fallback. This let's you do cool stuff like easily simulate swaps LIVE on your laptop. What I would love to see is node providers to offer something like this as a subscription, It would be a massive upgrade over just rpc services.
https://t.co/vYIhDnIoYI
[4]
Part of the inspiration was this video, which explores agents going “rogue” in a semi realistic environment:
https://t.co/Go004ABGTd
We’ll see how it goes.
🧵[1]
A slightly weird experiment I’m running:
AI agents have already found usable MEV alpha for us with me guiding the research.
Now I want to see how much of me is actually necessary.
I’m giving one our research history + code, $100 for gas/compute and a VPS.
[3]
I’ve been running research agents for a while now, and some of that research has turned into actual live alpha.
The next step would be to fork the entire setup across multiple servers and keep iterating on the most successful agents, essentially a natural selection process.