Robin Fritsch

Looking forward to presenting our paper "MEV Capture Through Time-Advantaged Arbitrage" (with @robin_ethz, @misilva73, @EdFelten, @convoluted_code) at MARBLE conference next week. Full program of the conference: marble-conference.org/programme-2025 The paper studies the value of TimeBoost coming from CEX-DEX arbitrage and (empirically) compares it to FCFS and PGA auctions.

@stonecoldpat0 Do I understand correctly that you're saying Timeboost is less extractive than PGAs? And if so, why?

Timeboost might end up being one of the biggest innovations overlooked by the MEV community who are laser focused on maximally extractive priority gas auctions. First few weeks exceeded my own expectations; very interested to see how it continues. Overview slide I made a few months ago to help explain the rapid auctions & that it only really facilitates back running.

@0x94305 Got it, that makes sense. Thanks! 👍

@robin_ethz I don’t think it actually matters how and if LPs hedge because hedging doesn’t increase *expected* returns. You can view the after fee LVR as the expected Fees-IL pnl too. And you are right that IL will be the same, but LPs collect less in fees in expectation.

Timeboost is worse for DEXes than it first seems. In this post, I'll explain (1) why Timeboost of 200ms is worse than a PGA over 200ms, and (2) why Timeboost is significantly worse than a PGA over the same time. These will be intricate points, so DO NOT expand unless you are an experienced LVR influencer. 1. Why is Timeboost worse for LPs than a comparable PGA? As you remember, LVR is determined by the average amount by which the CEX price overshoots the best bid or ask between arbitrage opportunities: For the price to overshoot the best ask, two things must happen: it must first reach the best ask and then (in the time remaining until the next block!) actually overshoot it. With a PGA over 200ms slots, the price will on average reach the best bid or ask sometime *in the middle* of a slot. This will always leave less than 200ms for the actual overshoot. (@MartinTassy and yours duly will hopefully quantify this effect precisely soon.) With a Timeboost of 200ms, the price will always have a full 200ms for the overshoot. All arbitrageurs except for the boosted one are slowed by 200ms, and it's optimal for the boosted one to wait for the full 200ms. I think this explains the result in the new paper by @kakia1989, Maria Ines Silva, @EdFelten, and @convoluted_code that arbitrageurs make more under TimeBoost than PGA: 2. Why is Timeboost significantly worse for LPs than a comparable PGA? But this is not the full story. One implicit assumption we made above is that when there is an arbitrage opportunity, the boosted arbitrageur (Bob) must take it after 200ms. And what happens if he doesn't? The idea seems to be that then some unboosted arbitrageur, Ursula, would arbitrage the pool instead. Should Bob be afraid of Ursula though? To win the opportunity, Ursula would have to submit her transaction immediately once an arbitrage opportunity appears. But she faces two complications: first, she knows that Bob might take the opportunity if the profit looks good 200ms later. So she is likely to get execution only if the price swings in the opposite direction and her trade is actually losing money. Second, when an arbitrage opportunity first appears, it's likely very small. So even if Ursula were to submit her transaction, it would be a very small one, and there would be no way for her to make it larger if the price continued to deviate in her favor. It's also unclear how much a threat of a small competing transaction would incentivize Bob to act faster. It feels that finding an equilibrium between Bob and Ursula in this game is not trivial, but it's clear that Bob shouldn't take arbitrage opportunities exactly 200ms after they appear. Long story short, to make Timeboost equivalent to a PGA in terms of LP returns, one would have to make the boost amount smaller than the PGA slot time.

@0x94305 Do you think this is a reasonable comparison? (In absolute terms, LPs end up with the same final position for path-independent AMMs.) (3/3)

@0x94305 And these profits can be viewed as LP losses, assuming LPs hedge each time an arbitrage trade occurs. However, this approach introduces different hedging strategies when comparing LP profitability between the two scenarios. (2/x)

In a paper: arxiv.org/abs/2410.10797 we look into effects of TimeBoost policy on arbitrageurs and MEV. Co-authors of the paper are: @robin_ethz, Maria Ines Silva, @EdFelten and @convoluted_code. Check out a research post about the paper here: research.arbitrum.io/t/mev-capture-…

@aklamun @ljfgudgeon Makes sense, thanks!

@robin_ethz @ljfgudgeon rETH tends to trade above peg and wstETH below peg for fundamental reasons (having to do with frictions of minting and redemption for each)

E-CLPs give better liquidity profiles for trading pairs. Here's how the liquidity profiles compare for this rETH/wstETH pair.

@aklamun @ljfgudgeon I see, thanks! What's the motivation behind choosing the price range asymmetrically?

@robin_ethz @ljfgudgeon It's a concentrated liquidity pool, so it sets a price range

@robinhanson Why do you have the impression that they do worse than we expect? I don't really have this impression

The question here is "Why don’t they then do better that we might otherwise expect", not "Why don’t they then do better than average."

Some people are really good at board games. Not just one or a few but they can do well at most any. Why don’t they then do better at life? How do board games differ so systematically?

@hdevalence It's a key feature of many AMMs, right? It enables passive liquidity provision

many AMMs attempt to create socialism for LPs unclear that this is actually desirable

As we have been reminded recently by the Compound DAO incident, governance attacks are a reality. What attack vectors are exploited most often, and which factors facilitate attacks? This is what we set to find out in our Systemization of Knowledge on Attacks on DAOs. (1/8)

@o_herminator Very promising team! :) Looking forward to your findings! Your McAMM ideas have been very influential!

Super stoked to research the MEV-L2 landscape. Looking forward to work with this excellent team!

@ballsyalchemist Super interesting plot! How do you define MEV Value here?

MEV categorization: Here are a rough % of each MEV types Atomimc arb ~ 40% (include various types of on-chain arbs) Sandwich ~ 25% CEX-DEX ~ 35% Totalling to around $1.3B+. There are still more classification work to be done on some long-tail and complex arbs, so this number could be slighly underestimating in terms of total MEV.

Thanks @modular_summit for organizing the event. And as always, amazing curation by @tzhen. Here are some highlight of what I presented this time⬇️ *Most of the graph and data I shared is based on data from Brontes, our open source MEV-tracing tool coming out soon.

@0xemperor Very cool work! 😎 Here, builder profits don't include any value they get from including their own transactions, right? @boez95

Very cool new data work, "Who Wins Ethereum Block Building Auctions and Why?", shows that - builders block market share a positive correlation w/ order flow diversity - positive correlation b/w market share & profit- features like exclusive signal, non-atomic arb, tg bot flow

@LiamKovatch @Polymarket And this one market/two market choice is separate from the AMM/LOB question. (You could potentially also have two AMMs for YES/USD and NO/USD.)

@LiamKovatch @Polymarket The difference: With one YES vs. NO market (be it AMM or LOB), passive LPs (that don't react) will lose everything, when the result becomes clear. With two market vs. USD, this will only happen in one of the two markets.

someday I will do a deep dive on our thinking re market structure @Polymarket, but there are nice properties of the orderbook that we get vs the cpmm. Perhaps most importantly is that makers can use the same usdc to collateralize orders across all books driving capital efficiency for the supply side. Additionally there are factors like the ability to support market-based idiosyncrasies like in-game taker delay. Empirically we've also seen the orderbook facilitate faster market formation from 1->n (slower 0->1). There are tradeoffs ofc and not always having a price is a big one for pms given informational angle. Let's connect - doing research on the next evolution rn cc @_loset

@0xdoug Polymarket used to use AMM pools. However, iirc they had pools with YES vs NO tokens with fees going into the pool, meaning passive LPs lose everything at revelation. Kleros seems to be giving AMMs another try: seer-2.gitbook.io/seer

Why isn’t there a strong AMM based prediction market platform? Much easier to attract passive LPs than market makers particularly for longer tail questions. Biggest barrier is setting swap fees relatively fairly for very volatile markets, but shouldn’t be that hard