Why Finance is Hard to Decentralize
The dream of cryptocurrency is to take all the overhead of the financial system—expensive buildings full of well-paid people making ad hoc decisions with all-too-human fallibility—and to replace it with zero marginal cost software that anyone can interact with and that functions safely and transparently without the need for any trusted party.
It hasn't played out this way just yet.
In one sense, financial systems are naturally decentralized already. The general structure in capitalist countries is that there are households and companies, both of which interact with the banking system and other financial intermediaries, and these intermediaries, especially the banks, are backstopped by a central bank. But there are many layers within that system; for example, a single transaction can involve a payment processing software provider, the issuing bank, the merchant's bank, and the credit card network. Every one of these roles has its established players, but in principle it's possible for a new player to enter at any time. A borrowing relationship like a mortgage is a similar mess of complexity: there's someone who gets paid to deliver a lead to an underwriter, then the underwriter generally sells that mortgage to some long-term holder, while every mortgage payment generates revenue for the mortgage servicing provider. And the holder of that mortgage might pay more fees or overhead to hedge their mortgage portfolio against changes in interest rates.
Annoyingly, all of these intermediaries have the role of a) standing between the person spending money and the person receiving it, and b) collecting a fee for the privilege of doing this. Clearly, on a one-time basis, it would be more cost-effective for the customer to transfer funds directly and save a bunch of fees.
On the other hand, the fact that these intermediary-heavy systems are continuously gaining share relative to their big privacy-first peer-to-peer competition—cash—indicates that there is some value being provided by intermediaries.
A quick look at a fairly common form of decentralized finance can show why this is the case. Take a DeFi protocol that, like many DeFi protocols, exists to lend money and pay out the interest to investors. It's a bank, without all the well-paid bankers who never seem to run into legal problems even when they cause financial problems!
The first problem you'll run into with a decentralized protocol is that it needs some kind of "oracle" to measure the value of collateral. If a decentralized protocol makes loans to small businesses, or to people buying houses, the likely borrower is someone who got rejected by a more conventional lender. Underwriting a business means trying to understand how it works, a process for which there are few standardized rules. Lending against real estate entails knowing something about the value of that real estate, and since real estate borrowers are often quite levered, the margin for error is small. But there is one category of lending where the oracle problem is easily solved: making margin loans against some product with a frequently-quoted price and reasonable trading volume. In the margin lending case, the value of the collateral can be measured based on the current market price, and the risk of that collateral getting impaired can be estimated by looking at the volatility of the asset.
Providing margin is economically equivalent to writing an out-of-the-money put option with a strike price where the liquidation starts. Normally, it provides a steady stream of income, but when prices move violently enough, the lender loses. And the faster prices move, the faster the lender's exposure rises: lend someone $5,000 against a $10,000 asset, and it's easy to ignore when the collateral's value falls to $9,999, but every $1 drop past $5,000 is $1 out of the lender's equity.
Worse than that, automatic liquidity sources change the way assets move. If there’s an incremental provider of leverage, that provider will a) keep increasing assets, and b) create an ever-bigger air pocket in between the price at which liquidations become a salient concern and where prices end up when they’re done.12
Suppose you identify this problem, and choose a solution: you’ll create a DeFi liquidity provider that accepts lower returns but only backs the safest loans. Letting customers borrow 95% of their purchase price is extremely risky, but letting them borrow 20%, even in volatile assets, is pretty safe. There are two issues here: first, such a DeFi product won’t be able to raise much money, since the marginal lender is not performing a very sophisticated analysis; at best, they’re trying to filter out the obvious Ponzi schemes, and at worst they’re just sorting by current yield.
Meanwhile, as these products get bigger, the classic Minsky dynamic takes hold: easily available credit increases the price of the assets that can be borrowed against, and decreases their volatility because someone can always lever up more to buy the dip.
Eventually, this system will unwind painfully; at any given instant, the most irresponsible borrower and the most irresponsible lender are the most likely to come out ahead, the borrower because their assets keep rising thanks to available credit, and the lender because they can always deploy more money at a return they find satisfactory. But if the market is increasingly defined by the dumbest participant, market prices will eventually get dumb themselves. And then all it takes is some minor perturbation to start a cascade of liquidations.
A margin lending algorithm can be built based on historical backtests, but what it can't backtest is the change in market structure caused by its own existence. This is by no means unique to decentralized finance, of course. It's a good description of what happened in 1987. Some smart academics discovered that an investor could replicate an options position using futures—as the market declines, selling more put options replicates the position that an options market-maker would have in order to hedge a put option, but in this case the market-maker doesn't have to get paid some premium for writing the option in the first place. This strategy got popular enough that when the market did face a big decline, it set off a wave of mostly-automated selling. The stock market crashed that day, with the S&P 500 down 20.5%. Futures crashed even worse, though, at one point trading at a 15% discount to the underlying stocks.3 The backtest for portfolio insurance didn't cover a period where portfolio insurance existed, and thus underestimated both the odds of a stock market crash and the odds that futures would crash harder, ruining the hedge.
Automated market-making, as DeFi proposes, has much the same problem. It's easy to create an automated market-making strategy, but this strategy is effectively a bet against volatility.4 In normal times, a decentralized market-maker will bumble along, churning out steady profits from the spread between bid and ask. And every once in a while, there will be a big liquidation or a burst of short-covering, and the market-maker will, by design, be automatically holding exactly the wrong position.
Automated payments can work—you can use Bitcoin, right now, to send arbitrary value to arbitrary individuals. But it works for a particular payments niche, with irreversible transactions and no built-in compliance.
That's a meaningful niche, both because international payments can be unusually clunky and because there's a population that wants assets whose seizure is technically difficult—some of these individuals are criminals; some are paranoid, libertarian, or both; some are nation states who don't always expect to be on America's good side. Being able to opt out of the dollar system, and then opt back in given good behavior or a change in the rules, has nonzero value. And since the dollar is the easiest-to-use aspect of sanctions enforcement, the addressable market for decentralized payments and decentralized stores of value is expanding over time. (No, Bitcoin is not a stable store of value, in the sense that its price fluctuates all the time. On the other hand, Russia's dollar-denominated assets actually experienced more volatility this year than Bitcoin did, in the sense that their value to Russia suddenly went straight to zero.)
Even if you did create a functional decentralized system, you’d face yet another barrier—decentralized systems are not easily discoverable. One of the first valuable businesses to be built on the Internet was Yahoo, which was given an apt backronym: Yet Another Hierarchical Officious Oracle. Start with a decentralized, permissionless system that anyone can use to create any content they want, and the first thing the market demands is some centralized authority to make sense of the mess. Bittorrent is a decentralized protocol, but has numerous centralized nodes that make it functional, including clients and and indexing sites like The Pirate Bay. Crypto, similarly, grows centralized onramps and offramps that necessarily tie it to the legacy financial systems it interacts with, such as Coinbase.
Subsets of finance, such as transactions, can be decentralized quite nicely. But the closer you get to the provision of credit—and, as it turns out, most forms of commerce involve some extension of credit—the more you're playing a game of adverse selection. Playing those games is very hard for the players who don't realize that's what they're doing.
So DeFi, broadly, will remain an impossible dream unless that underwriting problem is properly solved. But finding the specific layers at which decentralization can happen is still a useful project. One thing we may find is a more centaur-like financial system, where there's heavy interoperability between automated components without a single owner, but also numerous chokepoints controlled by specific individuals who can make judgment calls, update models and heuristics, and make the system legible to law enforcement and the courts.5
Perhaps the best way to look at DeFi is that it's a way to reach a new local maximum. The financial system has been on a long evolution from completely manual processes to mostly-automated ones that still start out by writing a computer program to do something a human being has been doing all along. This describes high-frequency trading, sure, but it also describes every financial innovation going back to the development of double-entry bookkeeping. The DeFi approach starts by aiming for full automation, but that means that if it evolves, it will evolve in the direction of finding the minimum required level of human intervention, instead of starting with maximal human intervention and finding the minimum acceptable level of automation instead.
Disclosure: I own Bitcoin, and Coinbase bonds.
A Word From Our Sponsors
The humble comma
The humble comma is — quite literally — what separates the world of data from falling into total chaos. Indeed, the Comma-Separated-Values file — or CSV — has become the workhorse of how most data is transferred between companies. But getting data from one company to another in a CSV is still a major painpoint.
Collecting, cleaning, interpreting and uploading data between companies safely and securely can take days, if not weeks. So Flatfile built a complete data exchange solution. The data magic: Flatfile enables you to move thousands of data records at lightning speed, time and time again by creating templates that can be infinitely customized to match the structure of any CSV file. Flatfile isn’t an app, it isn’t a service and it isn’t an importer. It is a fully customizable engine for data exchange that you can use in-house, in the cloud, and share with your business partners.
Why They Don't Drill
This is a good paper from commodities investor Goehring & Rozencwajg. One point it makes is that while the unit economics of drilling for oil are attractive right now, the economics of spending that same sum on buying back stock are even more appealing. Low multiples are both a tax on capital expenditure (each incremental dollar of capex produces less market value than it otherwise would) and a subsidy to buybacks (if the stock is trading at a discount to its fair valuation, buybacks make it cheaper).
A very off-the-wall policy proposal for fixing this would be to expand the strategic petroleum reserve's mandate, and allow it to not just buy physical oil but to buy huge amounts of stock in oil companies, too. The SPR is a special kind of investor with an unusual utility function: their mandate is cheap oil (at least at times when it's expensive), and one way to indirectly produce it is to ensure that the substitute good of oil-related equities isn't so competitively cheap. In the event that oil prices rise, owning those equities is a hedge against refilling the SPR on less advantageous terms. And if oil stocks underperform, it also means the SPR doesn't need as much money. The current value of the SPR is about $30bn, but it's been worth upwards of $100bn in the past, and $100bn is enough to buy 10% of every company in the S&P 500 energy index, which would probably have a market impact.
T. Boone Pickens once joked that the cheapest place to drill for oil was on the floor of the New York Stock Exchange. If that's the case, organizations committed to a stable and cheap supply of oil should adjust incentives accordingly.
Lots of people turn out to have known all along that FTX was up to no good and would inevitably collapse, but most of these people are too modest to share anything about when they shorted FTT, the exchange's token. The Information has tracked down some crypto traders who actually did this ($). The straightforward response to seeing a mispriced asset is to trade immediately, but in a case like FTX, that wouldn't work; even if something is unsustainable, it can keep reaching newly-unsustainable highs for a long time before there's a catalyst that kills it. Instead, traders have to treat their variant view as a synthetic option ($): not as an immediate trading catalyst, but as a way to have a variant view on the importance of specific news events. 2008 was a somewhat unusual case where the CDS market, which had the most advantageous pricing, got fairly illiquid early on; in many other short-selling case studies, the big risk-adjusted returns come from shorting after some of the bad news comes out. If you're confident an asset is going to zero, it's better to short when it's down 50% and you believe it'll collapse within weeks than to short earlier and wait months or years instead.6
Block.one and its founder have bought 7.5% and 9.3%, respectively, of Silvergate Capital. Silvergate's franchise as the most crypto-friendly bank has given it a rollercoaster ride recently; it went public in 2019 at $12/share, peaked at over $200 about a year ago, and most recently traded at $29. Silvergate's model is an interesting one; they seem to buy into the idea that everyone has a finite number of weirdness points that need to be spent strategically, so they've kept the balance sheet relatively conventional while getting deposits from, and providing financial services to, crypto companies. That makes them prone to overshooting in both directions. But it's also a systemically important institution for the crypto world, so crypto-adjacent investors have reasons to get involved that extend beyond buying a profitable bank at less than tangible book value.
It's not uncommon for pure e-commerce companies to venture into brick-and-mortar, and physical-first stores have gotten much more digital. But this kind of convergence shows up in other places, too; restaurants and grocery stores increasingly compete in the prepared foods market ($, WSJ). If a customer's job-to-be-done is that they want a cheap and convenient meal they'll eat at home, fast food and grocery are in basically the same market. Their cost structures are different, and the overall basket they sell differs, too (a restaurant and a grocery store might get the same gross margin contribution from add-ons, though in the grocery store case it would be several miscellaneous low-margin add-ons while the restaurant might just sell a high-margin soda). While restaurants and grocery stores have very different overall margins, they're much closer in structure for this kind of competition. Grocers, for example, have relatively lower labor expenses overall, but for prepared foods their labor input will be fairly similar to that of a restaurant. Ultimately, this story is a good reminder that while ancillary services can be a nice way to grow revenue, they're close to zero-sum overall.
Crypto exchange Binance has launched a proof of reserves website explaining how they measure customers' funds and revealing that, at least for Bitcoin, they have reserves fully covered. This is an interesting case where crypto and regulated finance offer different solutions to the same problem. There isn't a plausible way for Chase to provide a proof-of-reserves calculation, but the regulatory structure around banking means that they effectively have social proof of this instead. A crypto company can't rely on merely being popular and well-known (we all saw how that went), but it does technically have the option to demonstrate solvency instead.
Companies in the Diff network are actively seeking talent! If you're interested in exploring growth opportunities at unique companies, please reach out. Some top current roles:
- A company that's building a new hyperlocal bricks-and-clicks business seeks a market launcher who can close deals with small businesses and handle rapid scaling. (If you've ever wished you could have launched markets for Airbnb or Uber/Lyft in the early days, this is your shot...)
- An investment company using AI to accelerate investment in esoteric asset classes is looking for senior ML talent to build internal tools for ingesting and analyzing data (Bay Area, remote also a possibility).
- A well funded early stage startup founded by two SpaceX engineers is building the software stack for hardware companies. They're seeking a frontend engineer who can build powerful visualization tools for monitoring real-world devices. (Los Angeles)
- A firm using machine learning to customize investments is looking for a data engineer. (NYC)
- A company bringing machine learning tools to everyone is looking for experienced ML engineers with strong product sense. (NYC)
If you’re looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.
We're especially interested in talking to companies interested in remote contractors with finance and operations experience. If your startup needs half of a Head of Strategy or Director of Finance, please reach out.
A naive model would assume that prices get affected by margin positions once margin calls start happening, but in practice the closer prices get to a key level, the more likely price movements are to be determined entirely by whether or not they hit that level. This doesn’t mean they reach some trigger level and go straight down, because levered holders are incentivized to borrow more in order to support the price if they believe there’s a short attack. This can suck up lots of liquidity fast. ↩
Note that neither of these numbers are measurable in advance. Crypto price floors are set by unlevered crypto maximalists who will step in to buy when prices are low. The ceiling is even harder to measure. This old interview with SBF includes the useful point that corporate finance sets rough boundaries to where stocks will trade—if a stock goes up 4x or drops 80% for now reason, there will be a secondary offering or buyback in short order. But Dogecoin doesn’t have a CFO aiming to maximize shareholder value, so its price can go wherever. ↩
This event has revealing cameo appearances for just about every major figure of late 20th century investing. George Soros, for example, lost heavily in the crash, and wasn't able to liquidate his positions during the day. A few days later, Soros singlehandedly caused another brief futures-specific crash by panic-liquidating out of fear that the drop would continue. Meanwhile, Edward Thorp attempted some relatively safe index arbitrage, betting that the gap between futures and stocks would close eventually, even if the crash made things dicey. And Buffett started accumulating Coca-Cola shares when they were cheap after the crash, although he waited a few months. Everyone, in other words, played true to type; momentum investors chased momentum, quants looked for pricing discrepancies, fundamental investors took their sweet time but ended up making more than anyone else. ↩
A good example of the temptations here: every so often, especially after hours, you'll see a small stock trade print at an absurd price; the stock was at $30, and 100 shares changed hands at $25 before prices snapped back. What happens in those cases is that someone placed a big market order and the far-off-the-market buy got hit. Seems like a nice way to pick up pocket change: write a simple algorithm that places small limit buy and sell orders 20% or more off the closing price, and every once in a while someone sloppy will trigger them. But what isn't visible on charts is when this strategy goes bad—when there's market-moving news after hours, the orders don't get updated, and the stock takes a millisecond-long pause at $25 before hitting $15. This can happen with thinly-traded stocks sometimes; there was a company whose CEO and CFO were charged by the SEC with creating fictitious earnings and understating their company's share count. Somehow, the stock was trading down only slightly after hours, but the next day it closed down 47%. ↩
In one of many instances where crypto helps elucidate what's going on in the traditional economy, crypto shows that "limited liability" is a misnomer. A decentralized and pseudonymous system is essentially an extreme form of limited liability, where there are no reputational or legal consequences to bad behavior. In the traditional financial world, everyone running a company is economically a general partner, since sufficiently big mistakes can cost them more in reputation, or in prison time, than they initially invested in the enterprise. ↩
There are two big tragedies of short selling. One is that shortings' returns are worst at times when investors are enthusiastic about garbage companies, which means short sellers often have lower assets under management when the opportunity set is at its peak. The other is that the more successful a short seller is, the more they become constrained by size. "Wait until it's down 50%/80%/90% because it's going to go down 100% soon enough" is hard to implement if the first leg down reduces the company's market cap to, say, $500m, and the investor in question is running a $1bn portfolio. At that point they have to choose which concentration risk they want to take: being the biggest short seller in a small stock, or having a portfolio with lots of variation in idea quality and timeliness. ↩