Exchange runs can be described as abnormal withdrawals following rumors increasing risk on deposited assets on an exchange. In such a process, there are two possible primary outcomes, either the exchange has no problem providing the demanded liquidity, and the exchange run disappears, or the exchange fails to provide the demanded liquidity, and the exchange run is reinforced, pushing the exchange into a death spiral. If the exchange is liquid (1:1 backed assets), the game matrix looks like in Figure 1, where there is no incentive to change strategy, and others’ decisions do not alter one's outcome.
If the exchange can not meet the withdrawal requests, then the users have a problem. From a game theoretic perspective, the decision to leave assets on an exchange or withdraw them has two Nash equilibriums if the exchange lacks the needed liquidity to cover all the withdrawals (no 1:1 backing), as seen in Figure 2. The first one happens when everyone leaves the assets at the exchange, then nobody is impacted, there is no bank run, and everyone is happy (if pnl is in green). If user A withdraws and user B does not, then user B might be affected by user’s A decision to withdraw, same if B withdraws and A does not. Therefore, the logical conclusion for individual users on rumors of an exchange run would be to withdraw the assets from the exchange, causing the second Nash equilibrium, where users A and B both withdraw and both are unaffected by the decision of the other user (this equilibrium exists until the exchange runs out of liquidity). This second equilibrium causes and fuels bank runs, as defined by Diamond and Dybvig (1983). Given this game theoretic setup, even a false rumor is enough to start the self-fulfilling prophecy of an exchange run.
Friday 4th of November 2022, rumors started spreading that Alameda Research could be insolvent. Given the leaked balance sheets and the rumors, Binance, which had received $2.1 billion FTT (FTX token) after its exit from FTX’s equity in 2021, decided to liquidate its FTT position as a pure risk management measure. Binance's CEO announced the decision, and the rumors continued to rise. Every experienced player in the space knew the close connections between FTX and Alameda. On Sunday, Alameda’s CEO denied the rumors and argued that the leaked balance sheet was a partial balance sheet showing only a tiny portion of their assets.
It was clear that liquidity would flow out from FTX, given the rumors, but this should not fuel an exchange run because everyone believed that the exchange had one-to-one reserves of the deposited assets. As a result, users who had deposits on FTX thought they would not be impacted by the decisions of the other users withdrawing. Once the narrative shifted, users started to withdraw, with FTX seeing up to half a billion withdrawals in one day, according to on-chain data. Before the rumors intensified, the withdrawal game was more like the one shown in Figure 1, but then the reality became more like in Figure 2. As users started withdrawing en masse, the equilibrium moved to the second Nash equilibrium, provoking the exchange liquidity drain and giving FTX no option but to stop withdrawals without previous notice.
As a user with assets deposited on an exchange, you can do two things with the deposited assets: leave them on the exchange or withdraw them. As an exchange holding the user's assets, two things can happen, either the exchange continues to work as specified, or the exchange goes bust. Now, what a user would lose for the two possible asset custody decisions and the two possible exchange states, the cost matrix could be mapped as in Figure 3 below.
If the user withdraws, the cost that he would have incurred, regardless of the state of the exchange, would have been the transaction costs (excluding further possible costs like opportunity costs for simplification). If the user does not withdraw, then his cost is subjected to the state of the exchange, if not bust = no cost, if bust = all lost.
After the FTX rumors had been around for a day, well-known crypto investors started arguing about the rumors of FTX being in trouble, many could not believe the accusations, and soon, it became mainstream to say that the probability of FTX being insolvent was “<1%” - lower than 1 percent, as everyone would not believe that one of the biggest exchanges becoming insolvent was a possibility.
Now, coming back to the cost matrix above, if one assigns probabilities to the two possible states of the exchange, either not bust or bust, one can approximate the expected cost of leaving the funds on the exchange vs. withdrawing them.
Figure 4 shows the expected cost of leaving the funds on the exchange (Red) and the cost of withdrawing the funds (Black) plotted against the probability of the exchange going bust. As seen in Figure 4, the expected cost of leaving funds on an exchange increases as the probability of bust increases. If the cost of withdrawing is lower than the expected cost of leaving the assets on the exchange, then the rational risk-averse decision would be to withdraw the funds, as there is no benefit from leaving the assets on the exchange over withdrawing them. If the expected cost of not withdrawing is lower than the cost of withdrawing the assets, then the rational decision would be to leave the funds on the exchange, as there is no benefit from withdrawing the assets over leaving them on the exchange.
Guiding yourself solely by the expected cost calculation is unreliable since there can be a considerable dislocation between what you think the risk is and the actual risk. Returning to the FTX example, while the market argued about an almost inexistent risk, the real risk was probably close to 100% default probability, as it was later disclosed. Secondly, risk in such situations changes so fast, that it makes no sense to calculate the actual exchange risk and compare it against your withdrawing costs, as by the end of the calculation, the exchange might be gone. This approach tries to depict that one should decide on the risk of the exchange and the exposure to it, and the costs of withdrawing.
Based on the short list of precedents in the space, if rumors start, the best reaction is to withdraw funds as fast as possible. As mere rumors start, the decision to withdraw is backed by the game theoretic approach and the individual expected cost matrix perspective, where the cost to hedge against exchange default is minimal compared to the high loss exposure. As a risk management measure, it is recommended to only have assets deposited on an exchange if they are in the process of being sold/bought or are needed as collateral for trading activity. There are decentralized alternatives to centralized exchanges (DEXes), which offer deep liquidity and leverage. Centralized exchanges can be used as on-/off-ramps into DeFi to avoid counterparty risk, but additional DeFi-specific risks must be considered in such cases.
The FTX/Alameda crisis has sparked a movement of exchanges, such as Binance, Okex, Bitfinex, KuCoin, and Crypto.com, to bring more transparency into the space by publishing their reserve addresses (proof-of-reserves). Time will tell if these moves are intended to avoid withdrawals and possible bank runs or if it really is an attempt to bring the missing transparency to centralized venues. By providing proof-of-reserves, exchanges are going in the right direction. Still, it is crucial to be transparent about both sides of the reserves, the reserves, and the claims against them, which is a step that has yet to be made but would bring DeFi-like transparency to centralized exchanges.
After this crisis, contagion effects are still possible, as the old saying goes, “buy the rumors, and sell the news” or take the rumors as accurate.
This post is intended for educational purposes only and does not constitute the provision of investment advice, and is not intended to do so. The author and Primate Capital specifically disclaim all liability for any direct, indirect, consequential, or other losses or damages arising from any reliance on this article. Trading cryptocurrencies, derivatives, and structured products may involve a high degree of risk and may not be appropriate for all investors. Under some market conditions, it may be impossible to liquidate a position. Investors may suffer substantial losses and even lose the entire amount of their investment.