March 12: The Day Crypto Market Structure Broke (Part 2)
Note: A few days ago I published a postmortem detailing how the crypto market structure broke on March 12. This post, Part 2, will explore the potential solutions to some of the systemic problems outlined in Part 1.
To recap, one of the core structural problems is that current blockchains—both Bitcoin and Ethereum—simply do not support enough transaction throughput to facilitate global trading across many venues in volatile environments. When volatility picks up, traders need to shuttle assets between exchanges to arb price discrepancies, and they need to do this quickly (low latency).
In this post, I’ll explore some of the potential solutions, both market structure reforms and tech upgrades. This essay assumes a foundational understanding of various scaling technologies and the current market structure.
Modifying Market Structure
CONSOLIDATING TRADING VENUES
Right now there are about 10 major centralized trading venues around the world. If trading consolidates into fewer venues, then that would mean that 1) there are fewer venues to arbitrage, and 2) less need to shuttle assets on chain.
However, this is not likely to occur in the foreseeable future. Instead, liquidity has actually become more fragmented over the last few years as crypto has proliferated. In the first half of 2017, there were 4-5 major exchanges. Today, there are more than 10, and a long tail of smaller exchanges. This has been a result of local politics and fiat regulations (especially in countries like South Korea that have strict fiat regulations), the rapid growth of derivatives exchanges, and the fundamental divide between regulated and unregulated venues.
Moreover, as crypto continues to grow around the world, some countries appear to be cementing their local market structures. For example, it’s very difficult to see anyone launch a new exchange in China that challenges Huobi or Okex given their connections to the regional and national governments. There is a different, although somewhat comparable, paradigm in South Korea. This ossification ensures that liquidity stays fragmented for the foreseeable future.
CLEARINGHOUSES AND PRIME BROKERS
Because liquidity is fragmented across so many venues, liquidity providers are forced to split up their capital across many venues. This creates basis risk, which in turns limits the amount of liquidity they can provide per USD of collateral on each venue. If liquidity providers can cross-margin positions across exchanges, and net out longs and shorts, they can mitigate basis risk and provide more liquidity per dollar of collateral on each exchange.
Clearinghouses are the solution. However, they manifest in different forms. The two major categories are custodial and non-custodial.
The simplest custodial clearinghouse is effectively a prime broker that deposits collateral across exchanges, and acts as a clearinghouse across all of its customers. If a single prime broker has sufficient collateral across all venues, traders can face it rather than all of the exchanges, and abstract basis risk from individual traders. While the prime broker may need to occasionally rebalance assets across venues, this should dramatically reduce the demand for blockspace and gas on Bitcoin and Ethereum, respectively.
Traditional prime brokers in other asset classes are large banks like JP Morgan. They are some of the only entities with the amount of capital necessary to act as a trustworthy prime broker that both exchanges and traders trust. It does not appear likely that any large banks will offer prime brokerage services in crypto anytime soon, although Tagomi continues to add an increasing number of prime brokerage services.
A non-custodial solution is something like X-Margin, which aims to provide non-custodial cross-margining across exchanges. While the technical details are beyond the scope of this essay, X-Margin effectively allows traders to maintain custody of their funds with an authorized 3rd-party custodian, such as Coinbase Custody, BitGo, or Anchorage, and provide exchanges cryptographic guarantees that the underlying collateral is available in case it needs to be liquidated. X-Margin nets out all of a traders’ longs and shorts, further increasing capital efficiency.
This approach is novel, and is only enabled by recent breakthroughs in cryptography, and is untested. However, it’s beautifully simple in its approach, and makes sense on a first principles basis. However, non-custodial approaches suffer from a notable drawback: they cannot offer leverage, whereas traditional prime brokers offer leverage as a core part of their product offerings. Moreover, prime brokers natively offer delayed clearing and settlement, a feature that traders from traditional markets have come to expect.
Because most traders expect both leverage and delayed clearing services from their prime brokers, traditional prime brokers have a meaningful advantage over non-custodial clearinghouses. However, it’s unclear how this will play out.
DISALLOWING EXTREME LEVERAGE
One common criticism of the current market structure is that the derivatives exchanges—specifically via perpetual swap contracts—offer too much instantaneous leverage. BitMEX pioneered this by offering 100x leverage, and FTX and Binance have since one-upped BitMEX by offering 101x and 125x respectively. For context, with 100x leverage, a ~0.7% move against the position will cause liquidation.
It’s unclear how many traders trade with this much leverage. In May of 2019, BitMEX disclosed that the average trader trades with ~25x leverage, but it’s unclear how this has evolved since.
Will the exchanges cut back? Probably not. BitMEX has a shark tank in the office. They take pride in being ruthless. Obviously FTX and Binance offer more leverage just to say they offer more. Coupled with the fact that these exchanges are generally unregulated and are spread across a variety of jurisdictions, it’s hard to see them disallowing actions that their customers demand.
Another way to mitigate these problems is to introduce circuit breakers. If venues simply stop processing trades for 15-60 minutes—while still allowing for deposits and withdrawals—that allows arbitrageurs time to shuttle assets between exchanges.
However, given how heterogeneous the major trading venues are across many dimensions— geography, customer segments, products traded, and more—it seems very unlikely that the exchanges will ever agree on a global circuit-breaker provision.
Huobi, one of the largest derivatives exchanges based in China, just announced a new feature that they call a circuit breaker. However, it’s not a circuit breaker in the strict sense of the definition. It’s just a mechanism to space out liquidations over a few thresholds, rather than liquidating users at one point in time.
RFQs FOR LARGE LIQUIDATIONS
Liquidations typically occur via indiscriminate market orders. Market makers do not know how large the liquidations will be in advance. As liquidations tick up, market makers naturally widen their spreads. This increases volatility.
Instead, exchanges could hold a programmatic auction for the collateral via Request For Quote (RFQ). This could take place over the course of a few seconds.
While market makers’ ability to participate would be limited by the collateral they have available on the exchange that’s conducting the auction, this would nonetheless reduce volatility. To understand why, consider the difference between placing a $30M market order for ETH and asking a professional trading desk like Cumberland, Galaxy, or Jump for a quote for $30M of ETH. Clearly the latter will result in better execution because those firms have specialized tools to intelligently consume liquidity across every venue around the world.
Moving to a programmatic RFQ model would likely achieve similar results for liquidations.
Implementing Tech Upgrades
Many of the smartest people working in crypto realize that in order for crypto to mature from a fringe idea to a mainstream asset class, the underlying technologies must support greater throughput and lower latency. There are a lot of approaches to achieving these properties, each with a unique set of trade-offs. Below, I’ll touch on each of those approaches with a specific focus on how they would perform during heightened periods of market volatility where capital needs to quickly flow between venues, and often in a somewhat unidirectional manner. By unidirectional capital flow, I mean that capital is likely to quickly flow to individual venues (because that venue has the widest price discrepancy), but is unlikely to quickly flow back.
The problem with the Lightning Network (LN) is that channels must be fully collateralized by both counterparties ahead of time, and pre-filling channels is capital intensive.
Moreover, it’s not clear LN would actually solve the underlying problems. While it’s not really feasible to measure precisely how many BTC transactions were just shuttling money between exchanges on March 12, it’s generally estimated that 15-30% of transactions are traders shuttling money between exchanges. On a day like March 12th, this percentage should be higher than normal. On March 12th, about 750,000 BTC were moved on chain. With 10 major venues, that’s about 100 channels that need to be collateralized in both directions. What economic reason do exchanges have to pre-commit tens of thousands of BTC to pre-funded channels, especially when those channels are unnecessary 99.9% of the time?
While LN would accelerate the first few arbitrages, arbitrageurs would exhaust channel liquidity quickly.
Side chains theoretically solve both the throughput and latency problems. However, despite the fact that people have been discussing side chains since ~2014, they’re still not being used to a meaningful degree.
Blockstream launched the Liquid sidechain more than 12 months ago. Today, it has less than 1,000 BTC, and fewer than 1,000 transactions per day. While some exchanges are integrated with it, a few of the major exchanges are not, including major ones like Coinbase, Kraken, Binance, and FTX.
Why aren’t exchanges supporting Liquid? While it’s impossible to know with 100% certainty, the most likely reason is that they don’t trust each other.
Because of the inherent limitations of Bitcoin script, it has thus far been impossible to produce a trust-minimized side chain. Instead, Liquid uses an 11/15 multi-sig, where the 11 parties are known entities, primarily exchanges. But... the exchanges don’t trust each other.
Why don’t they trust each other? Because in the event of extreme volatility and market dislocations, it becomes rational for 11 of the exchanges to collude and steal BTC from other exchanges. The exchange business is brutally competitive, and given the fact that the exchanges operate across so many jurisdictions, the victim(s) would likely not have recourse.
Until researchers figure out how to develop a trust-minimized sidechain for Bitcoin, I’m skeptical of broad-based side chain adoption for BTC. And while it is theoretically possible to use Keep’s tBTC along with some of Ethereum’s more advanced scaling solutions, that introduces a lot of compounding technical risk. For example, if tBTC is flowing around on a low-latency Skale chain, how long do exchanges need to wait before accepting the compounding technical risk of tBTC and Skale (as opposed to a native Bitcoin transaction)? Given the lack of adoption of Lightning Network by exchanges thus far, it’s going to be years before exchanges underwrite these risks.
There are lots of marginal improvements available, including batching, and upcoming improvements such as Schnorr signatures and Taproot. However, during a period of high volatility, batching is not an option because traders want immediate withdrawals. This also limits the efficacy of gains as a result of Schnorr and Taproot.
These are the most commonly discussed avenues for scaling Bitcoin. Next, I’ll turn to Ethereum and smart contract platforms more broadly.
While sharding theoretically scales infinitely, in practice it will not. The core problem with sharding is that each cross-shard transaction requires one transaction on both the sending and receiving shards. Therefore, as the percentage of transactions that are cross-shard approach 100%, the global throughput of the system approaches that of a single shard. Furthermore, cross-shard transactions increase latency. This is problematic for a few reasons:
If all the exchanges just use a single shard in order to avoid cross-shard transactions...that is the same as the status quo.
If all the exchanges use different shards, then a huge percentage of transactions are cross-shard, limiting the benefits of sharding.
In an ideal world, each exchange would place ~1% of traders’ deposit accounts on 100 separate shards, and as traders shuttle assets around, exchanges would move traders to deposit addresses that are also on the same shard. Given how little the exchanges coordinate today (because exchanges are directly competitive, and generally do not cooperate), and that this kind of a system requires substantially more coordination than the Lightning Network or Liquid, I’m doubtful something like this is feasible.
Practically, the likely outcome is somewhere between the 2nd and 3rd scenarios described above. But it’s still unclear how any of this will actually work in practice, and what the second- and third-order consequences are. It’s going to take years after sharding is out to really model these flows correctly, and figure out if and how they can be optimized further.
These challenges are further compounded by the growth of DeFi. As liquidity pools in protocols like Uniswap, Stableswap, Shell Protocol, FutureSwap, and others grow, they will either centralize on a single shard (lower throughput), or fragment across many shards (lower liquidity per pool). The same is true of money markets like Compound and Lendf.me. As these pools fragment and users initiate multiple transactions across many shards to do what they would otherwise do in a single transaction today, the practical benefits of sharding become murky.
Note that the sharding framework applies to Cosmos Zones and Polkadot Parachains.
The vanilla optimistic rollups (OR) design only addresses the problem of throughput, and not of latency, because OR still relies on Ethereum Layer 1.
Furthermore, in order to even increase aggregate throughput, all of the exchanges need to agree to move into a single OR chain. If each exchange manages its own OR chain, the net result is likely to be comparable to a sharded chain in which 100% of transactions are cross-shard transactions because traders have to withdraw from one OR chain and send to another. There may be market makers who try to bridge OR chains, but they will need to create transactions on both OR chains, and because of the 1-2 week withdrawal delay in OR, will quote wide spreads due to high cost of capital (market makers do not want to speculate on the price of ETH-USD, and so will borrow ETH to provide liquidity).
However, there are variations of the vanilla OR design with additional improvements. With Skale’s BLS roll up, for example, latency is reduced to ~1 second. Using Matter Labs’ zk sync, traders receive instant cryptographic confirmation of transactions in the OR chain that settle onto the main chain later. However, even these solutions don’t address the problem of bridging shards and/or OR chains.
Therefore, unless all the exchanges agree on a single design OR framework, it’s unlikely to see this approach solve the core problem. As noted above in the sharding section, I’m skeptical that this will occur.
SCALING LAYER 1
The beauty of the Bitcoin and Ethereum models is that they expose a single, higher-level abstraction—with a single trust model—for application developers. Bitcoin and Ethereum are just a single database that apps can read and write from and to. That’s it. Apps don’t worry about any other abstractions like channels, new liquidity constraints, new trust models, or additional latency challenges. Apps just read and write from a single source of truth. It’s a beautifully simple abstraction, and one that every Web2 developer understands.
Solana is the only team building a new Layer 1 chain that’s hyper focused on scaling Layer 1 with thousands of concurrent validators. Today, it currently processes upwards of 50,000 transactions per second with around 400ms block times.
The problem that Solana faces isn’t technical, but social. It’s a new chain, and therefore will take a few years to earn the community’s trust and prove itself in the real world. Existing exchanges and DeFi protocols are rightfully hesitant to switch. But, as existing Layer 1 networks continue to falter under load—as they did on March 12—I expect that developers who are thinking about building web-scale experiences will begin to look to other chains that are purpose-built for performance.
Scaling blockchains is a multifaceted challenge. There are many market and technical approaches, each with unique trade offs. I am increasingly skeptical of theoretical approaches that have been known for years, but unproven, and am instead focused on newer approaches, both in terms of the development of market infrastructure such as prime brokerage, and in new technology such as optimistic rollups and new Layer 1’s.
Although it’s clear that the core problems will not be solved in the near future either via market structure or technical changes, I don’t expect another March 12th like event in the near term. Why? Simply because the move was so large and so fast that it wiped out a ton of leverage. There is a lot less leverage in the system now, reducing the risk of cascading liquidations and market dislocations.
However, until these problems are addressed, the same thing can—and probably will—happen again.
Inertia is potent, and the current market structure is complex and difficult to change. But existing market participants recognize the limitations of the current system—the system will break again when stressed—and therefore they will push the system to evolve as better options become available.
Disclosure: Multicoin Capital has invested in Tagomi, Skale, Solana, and Keep.