Research Summary: A DeFi Bank Run: Iron Finance, IRON Stablecoin, and the Fall of TITAN

Mechanism Design and Game Theory
1

TLDR

  • Stablecoins are designed to trade at a fixed exchange rate to US dollar. They are liabilities of the issuing protocol (like bank deposits) and thus must be backed by assets.
  • Algorithmic stablecoins like Iron Finance’s IRON use endogenous assets (e.g. TITAN) that are issued by the protocol with or without exogenous assets (e.g. USDC), making them fragile and more exposed to a potential asset-liability mismatch and run risks that can result in a self-reinforcing downward spiral.
  • We show that Iron Finance stablecoin fails when the protocol suffers from a large withdrawal shock. This can happen to all algorithmic stablecoins.

Citation

Saengchote, Kanis. “A DeFi Bank Run: Iron Finance, IRON Stablecoin, and the Fall of TITAN.” Puey Ungphakorn Institute for Economic Research discussion paper No. 155 (2021).

Core Research Question

Can an algorithmic stablecoin fail from a large withdrawal shock?

Background

  • Stablecoin: A fungible token (e.g. ERC-20) designed to trade at a fixed exchange rate based on some numerical value e.g. USD. There are many ways to instill trust among buyers and sellers, but they are mostly based on the ability to redeem for something with equivalent value.
  • Iron Finance: A DeFi protocol that issues two ERC-20 tokens on the Polygon blockchain: TITAN and IRON. TITAN is issued as a rewards, and IRON is minted (bought from the protocol) when participants send $1 worth of USDC (another stablecoin) and TITAN to the protocol and can be burned (sold to the protocol) for $1 worth of USDC and newly issued TITAN. Thus, IRON is a dollar-denominated stablecoin, and its value is “algorithmically” secured by the burning and minting of TITAN.
  • Bank run: a situation in which a financial institution that has liquid liabilities (e.g. deposits) cannot meet redemption demands because of mismatched assets, causing a withdrawal spiral and ultimately its failure.

Summary

  • Algorithmic stablecoins provide a more “capital efficient” way of creating stablecoins in a decentralized manner (that is, without relying on external, off-chain assets) as they do not require as much collateral as other models. Stablecoins also rely on algorithms and participants’ actions to arbitrage prices to the intended peg.
  • Participants react to arbitrage opportunities in a way that restores peg parity, but the mechanism can fail when the system suffers from a large shock in the shape of sudden, large liquidity withdrawal.
  • Thus, algorithm stablecoins are fragile because of their reliance on the value of endogenous tokens, creating a potential asset-liability mismatch risk.

Method

  • This paper visualizes granular transaction data obtained directly from the Polygon blockchain to show how participants react during the critical hours of the Iron Finance “bank run”.
  • Detailed data allows investigation of whether participants behave according to the mechanism design of the protocol. When arbitrage opportunities exist, do participants respond to them? Ordinary least square (OLS) regressions are used to detect the relationship between arbitrage opportunities (taking into account transaction costs such as swap fees but excluding blockchain gas fees, which are very low for the Polygon blockchain) and participants’ behavior at 10-minute intervals.

Results

  • Participants interact with the protocol as designed, reacting to arbitrage opportunities under normal times. When IRON trades below $1, participants buy IRON to profitably redeem for TITAN, which can be sold for profit.
  • However, if the prices of both IRON and TITAN are falling (triggered by a sudden, large removal of liquidity), the protocol can enter a downward spiral similar to a self-fulfilling panic. As IRON is arbitraged, more TITAN is minted which exerts further downward pressure on TITAN price.
  • Participants who observe this situation unfold may have lost confidence in the protocol and further withdraw their liquidity, turning this into a self-fulfilling panic like the classic bank run.
  • The price of IRON settles around $0.75, close to the value of the USDC reserves available per IRON. While it is technically possible to buy IRON for $0.75 for arbitrage profits, participants seem disinterested.

image
image1493×995 229 KB

Discussion and Key Takeaways

  • The financial system is based on trust, some of which can be automated using computer code. Here, computer code guarantees that 1 IRON is redeemable for $1 of USDC and TITAN, but the mechanism fails to restabilize. There are four important key takeaways here.
  • First, it might be because when TITAN price rapidly declines to unfamiliar territory – from $64.32 to 6 x 10-8, participants may choose to abandon (flight to safety) and sell both TITAN and IRON rather than arbitrage.
  • Second, TITAN is traded in limited places (mostly on-chain swap pools and not on centralized exchanges). Also, participants must swap USDC for IRON in one of the 2 USDC-IRON pools (other choices are TITAN-IRON and IRON-WMATIC, but not as deep as the stablecoin pools), but the same pools are also exit routes as they swap IRON for USDC. This lack of “exit liquidity” may be part of what accelerates the downfall.
  • Third, when participants lose their trust in TITAN, IRON collapses to the value of USDC. Stablecoins rely on trust in their redemption values. USDC is exogenous to Iron Finance, hence does not suffer the same demise. However, TITAN is endogenous and thus exposes the protocol to fragility. The loss of trust in Iron Finance is exemplified by the sharp drop in the price of TITAN is directly transmitted to IRON.
  • Fourth, Iron Finance is 100% on-chain, so everything is open to see. There is research in experimental economics which shows that bank runs can be exacerbated when participants can observe the actions of one another. In an off-chain setting, participants may hear many rumors of why a bank may fail but not see evidence of withdrawal. In an on-chain setting, participants still hear rumors but can directly observe withdrawals, which can hasten their actions.

Implications and Follow-Ups

  • Algorithmic stablecoins are fragile by design and relies on trust in the value of the endogenous assets.
  • Trust may or may not be automatable with computer codes. Further studies can shed light on which circumstances “code is law” is sufficient.

Applicability

  • Banks can be viewed as financial entities that issue debt-like liabilities (deposits) to finance mismatched assets (loans). A stablecoin issuer can thus be viewed as a “shadow” bank. If assets and liabilities are mismatched, a sudden withdrawal can trigger a run.
  • For algorithmic stablecoins, the assets backing the issued liabilities (stablecoins) are more precarious because they comprise endogenous assets with or without exogenous assets. The greater reliance on endogenous assets, the more likely a self-reinforcing spiral can occur, and the lower the “floor” price of the stablecoin.
9 Likes
2

After doing this study do you fundamentally agree with the conclusions of “Taming wildcat stablecoins?”

I’m sure @Astrid_CH could explain the piece better than I could, but if I recall correctly they were saying that stablecoins were neither money nor banks by legal definition, and suggested that the government could:

"(1) transform stablecoins into the equivalent of public money by (a) requiring stablecoins to be issued through FDIC insured banks or (b) requiring stablecoins to be backed one-for-one with Treasuries or reserves at the central bank; or (2) introduce a central bank digital currency and tax private stablecoins out of existence.” (Gorton & Zhang, 2021)

There were some pretty poignant objections to “taming wildcat stablecoins” though.

3 Likes
3

Thank you for the excellent summary @ajarnpai!

The demise of IRON and UST paints a painful lesson in mechanism design, especially as it relates to arbitrage incentives. It seems like once the trust in the project’s ability to defend the peg is eroded, arbitrageurs deem it safer to unwind their positions altogether. This, in turn, accelerates the downward spiral and creates a negative feedback loop.

Figure 3 shows that at around 6/17/21 02:00 the peg was somewhat restored – any theories as to why? Was it because redemptions had ended and all participants that wanted out were able to unwind their positions?

3 Likes
4

@ajarnpai Thanks for an informative summary of a very interesting paper!

In my opinion, your paper and its results feed into a larger discussion regarding the role of trust in decentralized finance. As you state in the paper, algorithmic stablecoins, which, by the way, still seems to lack a general definition, must by construction rely on some sort of trust mechanism. Personally, I agree with this view, and as far as I can tell this also constitutes the scientific consensus on the matter.

For instance, Ariah Klages-Mundt and Andreea Minca find in their recent paper (In)Stability for the Blockchain: Deleveraging Spirals and Stablecoin Attacks that under-collateralized, i.e. algorithmic, stablecoins can only be interpreted within their model with some variable describing user faith in the stablecoin system. This leads them to conclude that “the stability of the (stablecoin) system ultimately still relies on how people perceive its value over time similarly to how perceived value of Ether changes”. This is in line with your findings.

Now, in traditional finance, this perceived value, or trust, is typically instilled by variations of government insurance, regulation and/or expectations of governmental rescue interventions of too-big-to-fail institutions. However, these pillars of trust do not exist in decentralized finance. Therefore, I am curious to hear whether you believe algorithmic stablecoins will exist in a long-term DeFi equilibrium?

Further, in the summary you write “trust may or may not be automatable with computer code”. Yet, in environments with economically rational actors it seems difficult to establish trust in a stable asset without either insurance, well-established identities or fully/over-collateralization. In my opinion, the most ambitious stablecoin project at this time is the Gyroscope protocol which uses a stratified reserve but nevertheless is 1-1 backed with exogenous collateral. Hence, again, I guess my overarching question is whether a long-term sustainable future actually is feasible for algorithmic stablecoins in general?

3 Likes
5

This line definitely reminds me of the LUNA/UST liquidation. On the surface, it best sums up how $60B was wiped out from the market in just over a week.

What could cause this sudden, large removal of liquidity? Whales selling their tokens? And what percentage of the market cap/available liquidity needs to be moved to result in user panic?

Classic bank run. FUD, panic sells which further hurts the protocol/stablecoin.

This is a potential attack vector IMO. A malicious actor interested in bringing down the protocol could take advantage of the limited LPs and launch an attack (such as described in the paper).

With limited exit liquidity and pools to redeem the supporting token, restabilizing the stablecoin after its depeg could prove difficult.

Is it possible to have a hybrid algorithmic stablecoin - one that combines the backing of another token and real-world assets ($$)? Maybe this practice could limit/eliminate the effects of liquidation spirals and increase trust. :thinking:

Final thought: Algorithmic stablecoins are innovative, but I don’t think they’re necessary or what the market needs at the moment.

6

Some insurance such as FDIC would greatly help the stablecoin space in the case of liquidations/depegging.

The question is; how far are they willing to go?

1 Like