Indiscreet Borrowing
In a space in which borrowing can have negative interest rates and simple savings-type products have annual interest rates much higher than that of any savings account, the expectation is that double-digit yields should be the norm.
For your average investor, negative interest rates on borrowing are an unusual phenomenon, as seen by the public’s incredulity when Denmark introduced negative interest rate mortgages. Regardless, money-market protocols, which are at the helm of the DeFi surge, have also introduced their own form of negative borrowing rates. In these money-market protocols (including Anchor), borrowers can borrow and receive yield simultaneously.
In Anchor, borrowing is incentivized by having the Distribution APR significantly higher than the Borrow APR (261.19% vs. 16.16% at the time of writing). So, if the Net Borrow APR is -245.02%, a borrower is being paid 10x the amount to take a loan than they would be to deposit funds. These high rates are rampant across the DeFi industry and are necessary, to an extent, to bootstrap the protocol in the early stages.
However, this leads to reckless borrowing without considering the potential risks under volatile market conditions. High usages of margin and leverage further exacerbate indiscriminate borrowing practices with no forethought for market conditions This article will explore a cursory analysis of the high volume of forced liquidations on Anchor that occurred between May 17th and the 31st.¹
In order to see the magnitude of the liquidations, we examine the amount of bLUNA liquidated. During this period, there were a total of 7,667 successful liquidation transactions with an average of ~862 bLUNA liquidated per transaction (indicated by the green line), heavily skewed by a few large liquidations.
One can immediately see that liquidations as a function of time appear to not be random and that there are ‘pillars’ that occur where a large number of liquidation transactions are executed. For visualization purposes, we can trim the data by removing liquidations of less than 1 bLUNA and then remove the outliers. Note that we only use this to better examine the aforementioned pattern near the ‘bases’ of the liquidation columns and that the full dataset is utilized for all quantitative analyses. The trimmed mean rises to ~90 bLUNA liquidated per transaction.
Looking through all the liquidation transactions in this period, we can extract both the number of bLUNA received and the amount of UST repaid by liquidators. Note that most liquidation transactions came from a few addresses, so without the loss of generality, we can assume that the premium rates were set to 30% to maximize liquidation profits.² Under this assumption, the ‘implied price’ of LUNA arising from these liquidations can be calculated.
The plot of the implied price (in green) and the oracle price of LUNA (in grey) is shown below. Both sets of data are fitted by a generalized additive model.
It’s not surprising to see that the pillar-like clustering of liquidations coincides with the large drops in the price of LUNA. Looking at both datasets, it seems as though the oracle data is simply a lagged version of the liquidation data set. This is a pretty significant indicator that liquidations may be playing a role in the price movements of LUNA itself. Examining the chart below shows that the percentage difference between these two prices tends to hover around 7~12%.
In total, around 39.48 million UST was repaid to claim roughly 6.61 million bLUNA, with the largest liquidation of approximately 154,000 bLUNA.³ The LUNA-bLUNA pair on Terraswap can be further examined to determine the realized evolution of the exchange rates.
In most cases, one would assume that the LUNA/bLUNA rate would not stray too far away from 1. As LUNA prices crashed, along with the overall market, the LUNA/bLUNA rates also destabilized, dropping as low as 0.50 LUNA per bLUNA in one anomalous transaction.
One way to understand the capital in/outflows during this time period is to look at how much bLUNA was moved into LUNA (presumably to sell/swap), as well as how many trades occurred in which users were buying bLUNA at a discount using LUNA.
A total of 12.78M bLUNA was traded for 11.89M LUNA on Terraswap, resulting in a loss of 890K LUNA in ‘paper wealth’.⁴ The average rate of this trade was 0.9045 LUNA per bLUNA (or 1.1056 bLUNA per LUNA) between May 17th and May 31st.
As evidenced by the rates chart above, rates were realistically much poorer during the period in which LUNA (and the general market) was experiencing a strong downturn. During the exact timeframe, on the other hand, 11.61 million LUNA was traded for 13.00 million bLUNA at a difference of 1.39 million LUNA.⁵ The average rate of this trade was 1.1197 bLUNA per LUNA (or 0.8930 LUNA per bLUNA).
Looking at more detailed data, we can see that there was a net inflow of around 500K locked bLUNA from LUNA, signaling the confidence of the greater Terra community in the protocol even during times when UST was slightly off peg and the prices of ecosystem tokens were sharply declining.
Conclusion
By synthesizing our discoveries, we can learn a few things.
The first is that the tranche-like layered structure of loans on Anchor can be seen by the liquidation and price data. During periods of cascading liquidations, price movement in the underlying asset can be weakly predicted. The differential in the movements of the two sets of price data, correspondingly reflected in the Terraswap LUNA/bLUNA rates, shows that liquidators, as well as individual investors, were willing to sell off their locked bLUNA to buy LUNA at a discount.
Liquidators would not have experienced any realized loss had the bLUNA been converted to LUNA and then sold on the spot on an exchange due to the 30% buffer provided by the liquidation premium on Anchor. Even a heuristic analysis of the Terraswap rates shows that except for a few unusual trades, all bLUNA to LUNA swaps were traded at a rate with a maximum of 30% spread, with many executions occurring at much better rates.
Seems too perfect a fit for this to be a coincidence?
The maximum liquidation premium offered to liquidators should have never been set this high. A back-of-the-envelope calculation can be done for this parameter by defining an acceptable trade volume and then considering the typical spreads on the pair plus the additional tolerance during market stress (or equivalently the maximum allowable LUNA gain for liquidators). This value likely lies in the range of 7~12%. By reducing profitability, cascading liquidations due to the selloff of the underlying discounted collateral occur at a much lower velocity and additionally gives borrowers more time to react.
Perhaps the silver lining during the last few weeks is that the Terra community’s faith in a robust system has been unwavering, with many users willing to take on systematic risk by buying locked bLUNA with their LUNA.
So what can the Anchor team and community do to better weather these downturns and keep funds safe? From a product point of view, tools to automatically reduce risk levels by repaying loans or adding more collateral to existing positions will help manage loan positions.
As Henry Markowitz, the father of modern portfolio theory once said:
“In choosing a portfolio, investors should seek broad diversification, […and that] markets inevitably fluctuate; and their portfolio should be such that they are willing to ride out the bad as well as the good times.”
Anchor as a protocol should take heed of these words and rapidly diversify the collateral options on Anchor — not only with Terra ecosystem assets, but also external ones.
Perhaps the mantra most familiar to risky investors is: ‘high risk, high returns’. Leverage extends this spectrum so that gains can be magnified with a relatively smaller amount of capital. However, as markets can rapidly turn against the investor, improper risk management can lead to cascading downfalls in asset prices. Using Anchor as an example, many users would provide their (b)LUNA as collateral to borrow UST, which was then used to buy LUNA.
This process is repeated ad infinitum so that distribution rewards in the form of ANC tokens can be maximized. If a user borrows at the same percentage of the collateral each time, the total leverage ratio approaches 1/(1-f), where f is the fraction of UST borrowed relative to the collateral at each borrowing stage.⁶
As borrowed funds are used to increase the base collateral value at each round, the effect of a sudden downward price shock to the collateral price is amplified. The takeaway here is to manage financial exposure properly and to have cash on hand to reduce the risk of adverse price changes.
Special thanks to Mohammad Najafi, Sarah Kim, Brian Curran, and Irene Lee.
[1]: We hand-wave much of the technical machinery. For a more technical analysis, contact stanford@terra.money
[2]: An empirical analysis of the transaction data does show that this is the case
[3]: More precisely, 39,482,318.885056 UST and 661,334.8795012 bLuna
[4]: 12,781,757.851498 bLuna and 11,892,099.347586 Luna
[5]: 11,609,492.045971 Luna and 12,999,216.240232 bLuna
[6]: Recall the Taylor expansion for 1/(1-x)
