Terra: A Last Word
The champion-turned-traitor of decentralized money continues making governance mistakes which any Terra successor must avoid.
Before we return to our core topic (staking high-potential new protocols), we would like to add a few parting thoughts on Terra and the state of the algostable industry.
Terra will find a successor. Even now, the industry needs decentralized money more than the industry hates unstable money.
Crypto needs algostables, backed by crypto collateral, to have real-world applicability. A crypto-economy running on the rails of USDC—the Trojan horse CBDC of the US Federal Reserve—is, for most crypto users, a betrayal of crypto’s core promises of decentralization and pseudonymity.
DAI and similar algostable models are great, but... The overcollateralized algostable (DAI) model, while a great standalone business and an extremely secure standard, can’t become the stablecoin standard of the crypto-economy, for the simple reason that you need ~1.5X of cryptoassets (in DAI’s case) to issue $X of DAI.
Any USDC competitor is fighting a ferocious network effect. Scale of adoption is crucial; drinking from the poisoned chalices of external capital (centralization) or embedded leverage (undercollateralization) are the only paths to critical network mass.
Because of decisions Terra2.0 has made, Terra (v2) can never credibly return to the algostable arena. A decentralized, more collateralized successor to Terra, backed by real-world debt, must be built, either on Terra v1, from scratch, or atop an algostable survivor, such as FRAX.
While we wish Terra’s re-launch the best, we believe its mistakes continue to blaze a trail for potential successor algostables to avoid. The obvious blunders to us, which the market will project onto any future algostable model unless smart contract law proactively decrees otherwise, are:
Grotesque insider favoritism: Treating LUNA buyers pre-delegation-halt unequally relative to pre-depeg LUNA holders.
Failure to decentralize reserve management: To this day, a full accounting of Terra’s BTC reserves has, shockingly, not been provided. We think the evidence strongly supports an explanation that’s not criminal, but is highly embarrassing to Kwon and Jump Crypto, and as such has been abandoned in deafening silence by all parties involved (Gemini, Binance, and all the LFG Multisig signers).
Off-chain assets: Kwon extracted enormous assets out of Terra during the LUNA run-up which remain unaccounted for.1a. You should be very angry about their operating expenses Every month, TFL cashed out tens of millions of dollars for salaries and operating expenses. This went into hard money like USD, BTC and SGD. At their peak, TFL's 70 employees were receiving $180,000,000 monthly.
These flows were poorly documented by the industry. Any future algostable will require a holistic view of protocol opex and capital disposals (off-chain assets). The industry is in dire need of a comprehensive, public Chainalysis review of Do Kwon wallets to fully understand how much wealth was extracted from the Terra perpetual-motion machine, to learn the appropriate lessons from from Terra as it builds “Terra with the right fixes.”
Undercollateralization is a necessary, but not insurmountable, original sin
Before Terra stole the limelight, the algostable of choice was MakerDAO’s DAI. Every $1 of DAI is backed by around $1.50 of vaulted crypto-assets (mainly ETH, wBTC, and USDC). If your vaulted assets fall below 1.X, your vault gets liquidated. Maker charges an exit tax (liquidation fee), does not lose money, and then burns the DAI you originally borrowed against the vaulted assets.
This construction is as close to riskless as you can get in stablecoins, but carries a high price: every DAI minted into circulation requires more than $1 DAI of vaulted assets. Therefore, your stablecoin relies on people allowing you to vault assets for nothing or close to nothing if you want to grow anywhere near the growth rate of the crypto-economy.
Maker is moving aggressively towards backing real-world assets as a creative end-run around this constraint, but even with very successful real-world asset onboarding, overcollateralization will remain a fundamental constraint on DAI (given that RWAs will be capped at 1/3 DAI in circulation, and a higher cap would create the same types of concerns around Maker’s asset quality which hang over USDT, ie, “how good is the asset side of your balance sheet, really?”).
DAI solves perfectly for peg stability. It doesn’t solve for real-world network effects, and thus has suffered from a very slow rate of adoption. DAI has gone for institutional adoption and has no discernible ambitions to be used in everyday consumer transactions.
For that lack of ambition, DAI stands tall, while UST lies six feet under. As Rune keks and Kwon turns in his grave, of us still foolishly hope for a crypto stablecoin that can underwrite everyday consumer transactions.
The downside of high collateralization
Terra was always criticized for reckless undercollateralization. But full collateralization (or overcollateralization) and building a network which also grows faster than the overall cryptoeconomy are mutually exclusive. Overcollateralization isn’t without its costs.
Any network that wants to scale from zero to one (call it UST, DAI, Facebook, Amazon, or Uber) faces an immediate problem: negative equity. The first users are very expensive to acquire, much more so than the revenue they bring in. As the network grows, its revenue per user goes up, its cost per incremental user goes down, and its brand gets more valuable. At some point in the future, revenues outgrow costs at an accelerating rate, but you need to cross a chasm of sunk costs (negative equity) to get there.
A currency like DAI, which requires $1.50 of vaulted crypto-assets for every $1.00 of DAI minted, has a maximum long-run growth rate of 1/1.50 * (rate of cryptoeconomy growth). That constrains your growth by the amount of assets which others are willing to give you for free, and potentially concedes the positive money network effect to less-collateralized models. DAI’sovercollateralization bet was clearly safer than UST’s degree of undercollateralization, but it still doesn’t feel like a safe long-term bet in the winner-take-most algostable market.
Kwon’s model, as a later stablecoin entrant prioritizing high growth, necessarily opted for very low collateralization. It turned out to be too greedy, but prioritizing growth over stability was not necessarily wrong. Kwon meanwhile proved very farsighted in many other fundamentals of network design.1
Terra, after several VC rounds, tiptoed at the Rubicon’s edge by turning to Anchor, a bank that paid 23-25% deposit rates per year, as its default fundraising mechanism.2 As long as Anchor deposit sizes didn’t exceed the market’s expected lifetime value divided by customer acquisition cost (LTV/CAC), i.e., individual Anchor deposits weren’t large, Anchor added value to the Terra protocol, even when it burned money (which wasn’t always the case).
Kwon finally crossed the ponzi Rubicon, first with the MIM Degenbox, and secondly (after correctly deprecating the Degenbox under withering community criticism) by encouraging institution-scale Anchor deposits. The latter finally swung Terra’s internal economy out of balance, a la Thailand in 1997. The influx of massive, idle capital into Anchor destroyed the LTV/CAC balance of marginal-Anchor-user-value-added to Terra and created a massive excess of idle deposits ready to exit.
Terra1.0 vs Terra1.1
Terra v1 had a simple balance sheet. (h/t to the must-read Luca Prosperi for the terrific Excalidraw app suggestion, and all his other commentary too.)
After Terra’s reflexive death spiral to near-zero in May 2021, Terra realized that collateralizing their protocol with one recursive, unbacked asset wasn’t a good idea, and non-death-spiral assets would dramatically improve the resiliency of the peg. Over the next 9 months, Terra went from “0% collateralized” (0% non-LUNA collateral) to “20% collateralized” (80% LUNA-collateralized, 20% exogenously collateralized).
Algostable lesson #1: Clarify post-reorg treatment in bankruptcy
Exactly how Terra would use its centralized exogenous reserves (BTC & bridge assets) vs. decentralized native reserves (LUNA) during a crisis was never formally established. The theory was that if the UST peg broke to $.98 or less, ie, a serious run on the bank was just getting underway, the BTC would be the second firebreak of peg defense. Based on seniority, the de-facto Terra capital stack during a crisis might look something like this:
This brings us back to Terra’s big mistake #1: Terra’s bankruptcy reorg has screwed intra-crisis LUNA buyers, in favor of prior LUNA owners who didn’t support the protocol during the crisis.
LUNA’s capital-issuance machine worked by issuing new LUNA tokens automatically as UST was sold and burned below its peg.
As a “death spiral LUNA buyer” you knew you had a high probability of losing everything, but at a bare minimum, if you bought USD $X of LUNA during the crash while the seigniorage mechanism was still functioning (100 tokens at $5 for example) it should be worth 20x as much in post-bankruptcy reorg as someone who bought 5 tokens at $100 before the crash and sat on his hands.
However, in reorg, Terra has chosen to apply a 71% haircut to people who bought LUNA from $50 down to $.01 (when mint-and-burn was halted) vs pre-”attack” LUNA holders. (35% of LUNA goes to LUNC pre-”attack” holders vs 10% to post-”attack”; 1-1/3.5 = 71%).
Even though the later buyers took far more marginal risk defending the protocol than the OGs did.
The point here isn’t to criticize Terra’s decisions in bankruptcy. Someone was going to be screwed. Post-hyperinflation LUNA buyers are logically less sticky and less valuable to the protocol than longtime users.
However, as long as Terra’s decisions set the post-algostable-bankruptcy precedent, future algostable models will be assumed to screw highest-marginal-risk equity buyers in reorg as well — which will destroy the risk symmetry required to to provision the next algostable during capital flight.
Which will destroy any incentive to buy the next algostable during any period of instability.
Which makes them uninvestable as an asset class.
Algostables will need a “living will” within the smart contract to clarify post-doomsday treatment, similar to what banks have today.
Algostable lesson #2: decentralized reserve management
Almost 2 weeks after the blowup, we still don’t have a full accounting of how Terra’s BTC reserves were used.
We know that Jump Crypto threw $287M-$1B defending the Terra peg.
Not arbitrage. Naked buying.
In real life, crypto-adjacent tradfi behemoths don’t throw hundreds of millions of dollars on a dumpster fire out of the goodness of their hearts. More likely, Jump demanded and was pledged a large tranche of Terra’s BTC reserves up front in return for defending the UST peg on Curve (h/t @HsakaTrades) and Binance 11 days ago …
…, didn’t disclose that fact to the world, and when that peg defense failed, Jump redeemed and liquidated the pledged BTC at par.
While Gemini / Binance would be happy to throw Kwon under the bus, they would still like to be on good terms with Jump, which is still the Tiger Global of crypto investing. So everyone is sitting in embarrassed silence, explaining nothing.
I don’t care about embarrassing the Multisig signers; I care about the high probability that off-chain rehypothecation of reserves misrepresented a critical piece of information from the community—the apparent strength of the LFG reserves.
This course of action, while not illegal, would reflect very poorly not just on Kwon, but also on the LFG Multisig owners. It would explain the total silence around the topic from all parties involved, and is consistent with the cynical judgment that “the reserve was exit liquidity for insiders” (which the insiders also never really denied). In the absence of additional evidence, it should be presumed to be true.
Extralegal rehypothecation of reserves must be banned in any future reserved algostable implementation: all transactions involving a reserve wallet must be completely visible, implemented on public chains, in a decentralized manner which clearly accretes value to the protocol, i.e. in this case, the BTC used in the defense must mathematically remit significantly more LUNA than was minted and sold to create the reserves in the first place. (And if that level is $.80 or $.70 UST/3Crv because the BTC buys took place at higher prices, that’s fine. The point is to be able to burn algostable bad debt profitably, not to instill a false sense of security.)
The defense of Terra’s peg (very likely) invoked fiat mechanisms which took seniority over blockchain ones.
The bottom line is that Terra V2, despite having the deepest institutional knowledge of running an undercollateralized algostable and the deepest battle scars, can’t be trusted to do it again. Their justification (“a premeditated attack on UST”) is completely unproven. In Anchor, Terrans dug their own grave—yet they cling to it as a justification to privilege themselves in Terra v2 (post-bankruptcy Terra v1). They didn’t need to pillage peg-defenders to incentivize themselves nicely for Terra v2.
I am open-minded about Terra v1. It has been decentralized. Its mint/burn mechanism will be restored after Terra v2 re-launches. It has attracted a new horde of memestonk traders who will be able to put more resources into it if it gets any traction. It will still have an algostable model, which, if it can be improved, is still a unique differentiator.
Terra v2 has … some dapps with a Cosmos fork, a fraction of Terra’s old user base, and a still-very-centralized, still-discredited leadership (but with Kwon in a very diluted role)?
Lesson #3: Debt helps defend the peg. Debt Is Good.
In January, another infamous algostable, Magic Internet Money (MIM), saw a run on the bank when its founders were outed for extremely unethical behavior. However, MIM had been shamelessly marketed as a degen leverage machine, and had a huge amount of borrowing.
As MIM’s peg idiosyncratically wavered, people who borrowed MIM on leverage were able to profitably unwind their degen trades, which created buying pressure on MIM and restored its peg. MIM has survived to the present day, despite having completely discredited management, respectable scale (>$3B at peak) and respectable distribution. MIM, from memory, never broke .97 as a result, even while MIM in circulation dropped 50%+ in the span of a week.
A protocol with balanced borrowing and saving can rebalance more easily during instability. A protocol grossly skewed towards “saving” (Terra/Anchor) cannot. It just sees explosive outflows at the first sign of trouble.
Lesson #4: TFL [Kwon]-to-LUNA-to-offramp flows
TFL’s enormous expenses combined with not-crazy compensation fed rumors that he extracted hundreds of millions, if not billions, of dollars in Terra over its lifetime. It’s vital that a public-minded crypto firm underwrite a comprehensive Chainalysis-style audit of Kwon- and TFL-connected wallets, including any interactions Kwon personally had with smart contracts like the Degenbox which operated at the expense of the Terra ecosystem.
Who is likely to succeed Terra?
The most plausible undercollateralized algostable successor to Terra on paper is Frax. We hope to write a feature about them in the not too distant future. (DMs/Staking Mondays slots are open, @SamKazemian!)
From a user standpoint, Frax is much more like DAI than UST. Frax has very few users (n wallets where FRAX bal > 0 is around 6,500) relative to its time in existence, no supporting ecosystem, and lives on the Ethereum blockchain. Frax’s smart contract is also much more similar to DAI’s, as a “bank in a box.” It has not endured the stress test of multiple billions of free float in hundreds of thousands of wallets puking at the same time. Whether such a protocol can function on Ethereum at the very high levels of stress that Terra dealt with on Cosmos, without completely breaking down, is an open question.
In the near to medium term, the field is wide open for a Terra redesign that’s flexible and moderately conservative (50-85% exogenous collateralization) with a decentralized reserve management system, a living will, transparency around protocol expenses, a fixed off-chain burn mechanism—and a protocol governance structure that holds itself accountable for its mistakes, rather than invoking a made-up ‘network attack’ narrative to justify extortionate privilege.
By using Proof of Stake years before it became conventional wisdom, Terra slashed transaction costs to levels below most fiat banking systems without apparent costs.
By utilizing the Cosmos Tendermint consensus, Terra picked easily the most trilemma-performant consensus mechanism at the time, which scaled extremely amazingly well during Terra’s implosion.
Kwon erred with zero collateralization, but a) he had no other choice in the beginning, and b) he survived the May 2021 death spiral which resulted, and was well on his way to fixing that problem when Terra died.
By taking “the iOS view” of building an ecosystem (highly centralized) over “the Android view” (pseudo-decentralized), Kwon—a former Apple engineer—built a series of apps which were, by crypto standards, shockingly user-friendly and well-integrated.
By using third-party market makers balancing the UST peg with oracles instead of a USDC Circle API, Terra created the first censorship-proof stablecoin, theoretically immune to sanction from the US government as long as its internal economy stayed in balance.
Anchor’s APY wasn’t the commonly quoted 19.5-20% per year. The confusion was because an Anchor developer didn’t know the difference between APR (19.5%) and APY (19.5% continuously compounded, or 24% annualized). Anchor’s incentives were even more egregious than commonly understood.