The following is a proposal for an ALM framework for MakerDAO. The ALM Project includes @monet-supply @Primoz and myself.

I want to thank @Andy_McCall @Allan_Pedersen for their comments.

Summary

• This ALM framework proposes to optimize the MakerDAO balance sheet for MKR token holders profitability while being subject to a set risk constraints
• MKR profitability is a slightly modified return on equity formula
• Two groups of constraints are: solvency and liquidity
• Solvency risks aim to ensure the backing of DAI
  • Asset risk is the fluctuation of asset prices
  • Market risk is linked to liquidation issues of crypto-loans
  • Interest-rate risk is related to fixed-term lending
  • Counterparty risk is the failure of a counterparty on its expectations
  • Credit risk is the inability to repay a loan from a borrower
• Liquidity risks aim to ensure the peg of DAI
  • Maturity risk control the mismatch of maturities on the asset and liability side
  • Shiftability risk is the failure to get liquidity that was expected from an asset
• The first draft of an ALM dashboard is proposed

Introduction

We define Asset-Liability Management (ALM) as the tools and processes for optimizing the balance sheet to maximize token holders’ profits while being constrained by a defined risk appetite divided as solvency and liquidity. This document provides a framework to define the ALM problem space. It does not provide a solution as that will depend on MakerDAO community risk appetite.

The MakerDAO balance sheet is the core focus of this framework. Below is presented a possible balance sheet structure for the end of 2022 (purely figurative).

The balance sheet is composed of assets and liabilities (+equity, i.e. Surplus Buffer). Assets are used of funds (i.e. investments), we can list the following categories (from the most liquid to the least liquid):

• Stablecoins: This includes the PSMs and are our liquidity reserves (the closest thing from a US Dollar) in case of a drop in DAI demand. It is the first line of defense of the peg. It doesn’t provide revenues.
• Money Markets: Those are DeFi Money Markets investments (like D3M 13 on Aave). It is very liquid as well as it can usually be converted into stablecoins quickly (the actual process is more complex). There is an upper limit as such investment is impacting the lending ecosystem. This provides a low return.
• Short-term bonds: This represents an off-chain investment in short-term bonds (government or corporate). While profitability is low to very low, it doesn’t impact on-chain interest rates. Liquidity is quite good and the position could be unwinded in a few days with very low to no losses.
• Crypto-backed loans are currently the bread and butter of MakerDAO (e.g. ETH-A and WBTC-A). While the rates can be changed at will, there is an expectation of low volatility in rates (for the flagship products at least). Liquidity is therefore not immediate if we want to keep customer trust. It is nevertheless still quite short-term (a few months). Profitability is average.
• Institutional Vaults 8 are the next MakerDAO product that will offer non-contractual stability for institutions in terms of interest rates (fluctuation limited to 1% every 6 months). This is therefore a quite illiquid product line but provides revenues stability with a low to average profitability.
• Real-World Assets 9 (RWA): Private credit investments both off-chain and on-chain. They can’t be liquidated quickly but are expected to provide superior returns and good visibility.

On the liability side, which provides the funding of the asset part, we have only two buckets so far:

• DAI: the MakerDAO stablecoin. We also include the DSR 2 in it. The cost of this capital source is very low (DSR is 1bps) but growing demand is quite difficult in terms of product and marketing and can’t be commanded (except by increasing significantly the DSR which leads to other issues). Some part of DAI issuance is stable and some is volatile (subjected to be converted to another stablecoin in case of stress).
• Surplus Buffer: The capital reserve of MakerDAO which belongs to MKR holders. While there is no cost of capital per se, it can be derived from the market valuation of MKR.

Tokenholder profitability defined

First, what do we mean by maximizing tokenholders profits?

ALM focus on the balance sheet, hence each balance sheet item will be assigned an annual percentage yield. If positive, the balance sheet item is producing profits. Most assets should produce revenues (e.g. loans will provide interest revenues). If negative, it is producing expenses. Those will be mainly liabilities (e.g. DAI Saving Rate where the protocol incurs an expense to reward DAI stackers).

Using return over equity metric

We suggest optimizing for Return Over Equity defined as follows. We weight each balance sheet item value (Vi for the value of balance sheet item i) per the annual yield (positive or negative) of this balance sheet item noted APYi. The sum of all those balance sheet items is then divided by the amount of equity of the protocol (i.e. the Surplus Buffer).

We can discuss incorporating direct costs to the APY of balance sheet items. Indeed, DAI holdings are due to integrations with the wider ecosystem and a brand built over time by marketing. While the APY cost of DAI is 0%, one can argue that marketing costs should be taken into account. On the asset side, Oracles costs are significant and some collateral facilities are at a loss if those costs are taken into account. Core Units costs in general are quite a significant expense. Nevertheless, most of those costs are fixed (or at least not scaling linearly) and might end up being subjective. While those considerations should be used to keep or dispose of a product line, it seems more relevant to stick with pure capital items for ALM purposes. The same goes for structural costs and exceptional items (i.e. liquidations income).

As of end December 2021, the balance sheet of MakerDAO is the following (values in $M)

This gives an annual return of $143M which, divided by the equity (i.e. Surplus Buffer so $64M), gives an ALM RoE of 222%.

Comparing RoE with Net Income

One consequence of using RoE over broader profitability metrics (like net income) is that it allows optimizing the amount in the Surplus Buffer as well. By construction, reducing the amount of equity is relutive to RoE if, and only if, the price of MKR ($2000 as of writing) is below the equity per MKR ($75 as of writing). The same applies on the reverse but the catch is to find usage for such additional capital. As discussed, the current RoE of MakerDAO is around 220% and it is unclear how to allocate additional capital without decreasing the RoE.

Comparing RoE with Return over Asset (RoA)

Another metric that can be used is Return over Asset (RoA). This is the average yield of the assets items.

The problem with the RoA metric is that it doesn’t take into account the leverage. Assuming a zero-cost funding, RoE = RoA * leverage. Using the number as end of December 2021, we get a RoA of 1.57%. This could largely be improved by decreasing the PSM, which means decreasing the utility of DAI. This is obviously not a sane course of action, therefore not a good metric.

Comparing RoE with Net Interest Margin (NIM)

The Net Interest Margin (NIM) is the weighted yield on balance sheet items divided by the total assets (the size of the balance sheet).

Same as the RoA, the NIM doesn’t take into account the inherent leverage of MakerDAO.

Now that the profitability objective is defined, what are the risks that constraint the balance sheet?

Comparing RoE with Economic Value Added (EVA)

Economic Value Added is a derivative of the Net Income (or NIM depending if we excluded non-balance sheet items revenues/expenses) where we subtract a capital charge corresponding to value that MKR holder expect.

It is better than the RoE in the sense that it favors increasing the capital base as long as we can generate value over the capital cost. Nevertheless, there is a great sensitivity to the capital cost input. Moreover, the metric is less easy to interpret.

Solvency

Solvency is the constraint of having assets being worth more than liabilities, i.e. DAI being backed. This constraint is built deep into the Maker Protocol. If at any point in time, there is insolvency the protocol will issue MKR token on the market to build up reserves (more precisely removing the “no longer backed” DAI from the market). Maker also has a concept of Surplus Buffer that provides first loss protection.

Solvency concerns can be expressed with the following chart.

• Expected losses (related to risk mostly) are statistically expected losses and are embedded in the pricing of each product. That means that, on an average year, the profits (after deducting the cost of funds) from a product are greater than expected losses. Otherwise, it wouldn’t make sense to lend in the first place. With the law of big numbers, we can produce strong statistical models.
• Unexpected losses happen when it is not possible (or not accurate) to use statistical models. The economic cycle can lead to systemic credit risk (e.g. subprime crisis where borrower default correlation goes to 1). Other sources of unexpected losses are market risk (e.g. ETH losing 90% of its value in a few minutes) or operational (e.g. zero-bid auctions Maker endured in 2020). Depending on the level of unexpected losses, 3 scenarios are possibles:
  • Losses < Surplus Buffer: In such a case, the bad debt is subtracted from the Surplus Buffer. While it decreases the share of the capital of the MKR holders, it doesn’t impact their share of future earnings.
  • If the Surplus Buffer is not enough to absorb the loss, then the system will sell MKR tokens on the market to fill the gap. This lead to MKR holder dilution which reduces their share of underlying capital (which is 0 at this stage anyway due to the previous step) and future profits. If we assume a $2B market cap for MKR and an unexpected loss of $1B, each pre-existing MKR holder will be diluted by 100%. There is a reflexive effect as an unexpected loss will impact negatively the MKR price which will lead to greater dilution, itself impacting the MKR price. Therefore, while it is unclear what level of unexpected losses is sustainable compared to the market capitalization, it is possibly an order of magnitude smaller than current market capitalization (e.g. a maximum $200M loss can be absorbed by a $2,000M market cap). Some early work used the DCF 3 to estimate the possible dilution.
  • If MKR market cap falls below the loss (after Surplus Buffer deduction), then the system is defaulting as it can’t raise enough DAI on the market by selling MKR.

MKR holders have to decide on the size of the Surplus Buffer which will decide the probability of MKR issuance to cover exceptional issues. As the price of MKR is outside the control of MKR holder, it will also define the probability of a protocol failure (but without providing a clear metric of the maximum sustainable loss).

On the solvency front, we can identify 5 main sources of solvency risk:

Asset risk

Asset risk is the risk of an asset fluctuating in the market. This could be a direct holding of the asset (like Fei protocol is doing ETH) in which case each decrease of the value of the asset is impacting the solvency (as of writing Fei would be insolvent if ETH drop by more than 60%). There is also a recent trend of DAO to DAO swaps where DAO exchange some governance tokens. This would create a market risk exposure to the tokens. So far, this risk is quite limited for MakerDAO as most assets and most liabilities are expressed in DAI. The exception is some ENS tokens in the DsPauseProxy contract, but the amount is small ($1.4M as of writing) and usually excluded from consideration anyway.

Asset risk is for fluctuation of the price but orderly markets, in case the risk is not so much asset fluctuation but sudden moves, it is labeled market risk which is the next risk.

Market Risk

For crypto-loans, the Risk CU provides a market-driven market risk estimation 7. It is the risk of capital losses using stressed scenarios (hence falling into the unexpected loss realms as no loss for crypto-loans is expected in normal circumstances). This metric uses market liquidity and a granular level of vault riskiness.

Therefore, Risk CU is providing a direct capital at risk metric that can be used to size the Surplus Buffer.

Interest-rate risk

Interest-rate risk is the potential impact, adverse or otherwise, on the net asset value of a financial institution’s balance sheet and earnings resulting from a change in interest rate.

A change of interest rate impacts the net present value of an asset that has a non-zero duration. Duration is a measure of the price sensitivity of an asset (debt instrument) with interest rates fluctuations. A duration of x means that the price of the debt instrument move by x% if interest rates are moving by x% (we are using the modified duration definition). Assuming low-interest rates (<5%) and a zero-coupon bond structure (what we use in DeFi usually), the duration is close to the maturity of the debt instrument. Therefore, a zero-coupon bond with a 3-month maturity has a 0.25 duration, a 1-year zero-coupon bond has a duration of 1.

At this stage, the mere definition of the market interest rate is almost impossible in DeFi. Is it ETH-A SF? The DSR? Is it money-market rates (Aave or Compound, with or without incentives)? Should we stick with the TradFi yield curve?

We suggest considering differently the two sources of interest-rate risk: Real-World Assets and Institutional Vault (IV).

For RWA, it should be possible to use the TradFi yield curve.

For institutional vaults, we suggest using the reference rate (for instance an ETH-based IV will be based on ETH-A SF). From there we will compute the loss that the “fixed” component of IV will generate. If we have a 1B institutional vault with an ability to increase rates only by 1% every 6 months, an increase of 1% of the ETH-A SF will lead to a “virtual” loss of $5M (virtual in the sense that the current Maker Protocol don’t account for that). The ETH-A SF, considered fixed going forward, will be used to actualize the loss if this is significant.

More elaborated models will be worked on when this topic will be of greater importance.

Counterparty risk

The counterparty risk is the possibility that a counterparty will not be able to meet its contractual obligations. We will treat failure to repay a loan as a separate risk (credit risk which is still a kind of counterparty risk but addressed specifically in the next section).

The counterparty risk, in this framework, should have a low probability of occurring but with a possible significant loss when it occurs (loss given default being often 100%). Being unpredictable, it will lead to an unexpected loss.

For crypto-loans, the Risk CU already monitors the different counterparty exposures 4. As we can see, the main one is exposure to Circle due to the USDC PSM. The second one is to BitGo due to the WBTC crypto-lending.

Counterparty risk should be kept reasonable by not allowing any single counterparty to be a significant risk to MakerDAO. Sadly, this is not possible yet. Therefore, some key counterparty risks should be listed in a whitelist. We can think of Circle, BitGo, and Paxos currently.

Credit risk

Credit risk is the possibility of a loss resulting from a borrower’s failure to repay a loan. Using this definition crypto-vaults users are not incurring a credit risk as we don’t expect them to be liable to repay (as we don’t know them in the first place).

Credit risk is decomposed by the probability of default (PD), the probability of occurrence of a credit event, and loss given default (LGD), the expected percentage of loss when a credit event happens. Another component is the expected exposure at default (EAD) which represents the amount at risk when the credit event occurs. It is understood that when providing a credit facility, the usage of it will increase just before the default. EAD allows taking that effect into account.

Therefore, credit risk is mainly in the realm of RWA and most parameters will be given by the RWA-related CUs.

Liquidity

In TradFi, the concept of liquidity is the ease with which assets can be converted into cash. Banks need to be able, at any time, to meet customer demand for deposit account withdrawal. MakerDAO doesn’t have the same issue as DAI are not redeemable for anything (only for $1 of any kind of asset at Emergency Shutdown). Nevertheless, DAI should be worth $1 to fulfill its promise of being a stablecoin. Stablecoin holders should be able to use the stablecoin to buy something worth $1 on-chain or off-chain at any time. Therefore, while there is no contractual liquidity risk, the market demand liquidity DAI (ability to buy/sell DAI for $1 at scale in secondary markets) to be taken care of by the DAO. The history of DAI at par value is studied in another paper 5.

The mitigant for this volatile kind of liability (stablecoin) is to keep liquid assets as reserves. Currently, MakerDAO keeps a huge amount of USDC (and USDP on a lower scale) as reserves as DAI demand is far above DAI supply from crypto-backed loans. DAI holders can get USDC or USDP without a spread (or DAI from USDC/USDP) and from there convert to US Dollar without slippage. This provides a risk-free arbitrage enabling a strong peg. As of writing, such stablecoin reserves represent >60% of the balance sheet which is seriously decreasing our profitability level and is unlikely to be an optimal situation. Liquidity should be strong to sustain any stress but is also costly for MKR holders. Hence the motive to study what is the optimal liquidity.

Holding stablecoins as reserves is not the only way to ensure sound liquidity. Influencing market behavior by moving interest rates on both our crypto-vaults (Stability Fees) and our DAI product (DSR) can decrease loans demand and increase DAI holding. Nevertheless, increasing DSR to source some DAI holdings might prove difficult in times of stress. Pushing our borrowers out by increasing their Stability Fees is neither a great outcome as it would put stress on our borrowers. This is not commercially great.

Short maturity assets can be a precious intermediary solution. MakerDAO is already providing capital on Money Markets through D3M (Direct Deposit DAI Module). While withdrawing out liquidity will increase the borrowing rate there, the commercial impact is lower as borrowers on money markets are not our direct customers. Still, there is a limit on how much MakerDAO can leverage Money Markets protocol as it could cannibalize its own crypto-loans product line. A TradFi solution would be to use money-markets or short-term bonds investments products to invest the excess liquidity. Real-world money markets and DeFi Money Markets are, at this stage, still uncorrelated. Therefore, scaling wouldn’t be an issue.

We will decompose the liquidity constraint into two parts. First is the maturity risk, i.e. the risk of using short-term funding to invest in long-term assets (again assuming DAI is short-term on a behavioral basis and not on a contractual basis). The extent to which this mismatch is acceptable is a parameter that should be set by the community. Then we will emphasize the shiftability risk which represents the inability, due to market condition or others, that a long-term asset expected to be used as a liquidity source is no longer redeemable, sellable at an expected price, or usable for a repo operation.

Maturity risk

When dealing with the maturity of the items of our balance sheet, using the more likely “contractual” maturity is quite deceptive. As already discussed, DAI has no maturity yet we need to keep the peg to remain an ongoing business. It is nevertheless overly pessimistic to assume that all DAI holders will try to get out of DAI at the same block.

While the ALM literature provides some solutions, none is applicable directly to a DeFi setting. For instance, the Basel III framework is using the type of deposit account (retail versus corporate, …) to issue a maturity profile (or more precisely a risk of run-off). Believe it or not, the run-off rate of a retail internet account is twice the run-off rate of a retail branch account. Basel III doesn’t provide any guidance regarding smart contracts or DAOs customers.

One strength of the blockchain is that, while customers are usually anonymous, their behavior is public (even if making sense of it is not easy). How to define the maturity of DAI remains an open question and will require more work.

The same applies to the crypto-loans which are most of the assets of MakerDAO. Those have no real maturity but we could “close” them by increasing the Liquidation Ratio (1000%+) to the maximum or using a punishing Stability Fee (100% APY). This is nevertheless not realistic if we want to be an ongoing business.

You can see below an illustration of MakerDAO current liquidity maturity profile (numbers are only for the illustration). Items on the positive side are MakerDAO assets distributed by maturity. Stablecoins can be sold within a block to meet DAI selling demand (through the PSMs) hence a 1-block maturity. Crypto-loans are distributed between the 1-week (announcement and first rise in rates) to 1-year (people being not rate sensitive). On the negative side, you have the liability items with the Surplus Buffer being at the 1-year+ maturity (actually having no maturity at all) and DAI that are distributed quite arbitrarily around the time horizons. The purple line represents the cumulative liquidity gap for each maturity. If positive, it means that there will be a deficit of asset funding. It will be solved by DAI being below par (speculators being willing to bet on the ability of MakerDAO to solve the balance sheet issue). As we can see, in this scenario MakerDAO has no liquidity maturity risk. This is not a surprise as we have a very strong liquidity buffer currently.

If we assume the development of RWA and Institutional Vaults (IV), hence a lower liquidity buffer, we could imagine a liquidity maturity profile showing a $1.4B funding deficit at the 1- week maturity.

As an aggregation of the liquidity maturity profile, we could use the Liquidity Coverage Ratio (LCR) which is the ratio between High Quality Liquid Assets and expected 1-month outflow outflows (i.e. net decrease in DAI holding) under a stressed scenario.

HQLA can be both assets that have a short term maturity (stablecoins which have a 1-block maturity or Money-markets instruments). We will use all assets with an expected maturity up-to 1-month.

Shiftability risk

As we have seen at the beginning of the liquidity section, one solution to ensure the peg is to keep liquidity reserves, in our case mainly in USDC. Having huge reserves of stablecoins (50% of the balance sheet as of writing) is nevertheless quite opposed to our aim to maximize the balance sheet profitability for tokenholders. The yield of holding a stablecoin is 0%.

Moreover, even USDC can be seen as just a proxy of what our customer wants (US Dollars). If the link between USDC and the US Dollar is severed, i.e. Circle suspending convertibility for any reason, those USDC would no longer perform the expected liquidity role for DAI. This would be an indirect shiftability risk, i.e. the risk of being able to convert an asset into liquidity. One solution would be to provide many kinds of liquidity reserves (which we have with USDP and GUSD). Indeed, the Maker Community should enforce diversification in the used liquidity assets.

Besides this low immediate shiftability risk (can USDC be converted to US Dollars?), shiftability risk makes more sense when using longer-term investments that are assumed to be liquid enough to meet liquidity needs. We will consider short-term investment and long-term investments separately.

On the short-term front, we can see money-market investments, both DeFi and TradFi.

DeFi money-market investments are the usage of D3M (Direct Deposit DAI Module) for Compound and Aave. Those money-markets work as a repurchase agreement where the DAI deposited can be lent by the money-market platform to third parties. The borrowing rate is sensitive to the level of liquidity in the pool which should force some level of liquidity. Nevertheless, there is no guarantee that DAI deposited could be redeemed within a fixed period of time. This problem grows as MakerDAO is a significant lender of a pool.

TradFi money-market would be represented by investment in money-market-like ETFs or money-market-like managed accounts. Shiftability in such a case would be more about the inability to get a good price when needed. There are some rare exceptions where shiftability doesn’t work for money markets or short-term bonds investments. We can recall the subprimes and the Covid crash 1. The rise of uncertainty in those periods make price uncertain and market-makers, facing the liquidity constrain themselves, are no longer there to provide an orderly market.

The key for the future DeFi system might be to manage shitability on long-term investments. The shiftability theory was introduced by Moulton as a new and unconformist (in 1918) way to increase the maturity mismatch of financial institutions. We will illustrate with two examples: one from TradFi and one from DeFi.

In TradFi a lot of financial entities are holding long-term assets (like Treasury bonds) but are funded short (e.g. from Money Markets). They don’t face the liquidity constraint because it’s quite easy to do repurchase operations (repo) with a Treasury Bonds. Therefore, even if you can’t/don’t want to sell the asset you can get funding from it. The yield curve being usually upward, the operation provides a profit. The corner case is when shiftability breaks and no repo counterparty can be found.

The equivalent in DeFi would be to own ETH by speculating that the price will increase over a long period of time. If MakerDAO would face a liquidity constraint (i.e. DAI != $1), it would be able to repo this ETH position on money-market like Aave or Compound. In case of a need of liquidity, such ETH position will be used on the money market to borrow other stablecoins to defend the peg.

While ETH is the best example today, it is likely that DeFi will evolve to provide more term yield-generating products that would provide sound investments and provide great shiftability as well (looking more like Treasury Bonds). Yet, here gain, shiftability works until there is great stress in the market which breaks shiftability (in the first sense or by providing punishing rates). It is too early to have a more detailed discussion, but this is an area to follow carefully.

ALM Reporting

If such a framework is accepted by the MakerDAO community, the next step will be to define the risk parameters and provide a monthly (and/or real-time) report to monitor if the MakerDAO balance sheet is within the limits that the community agreed on. This would be useful to warn and take action.

While the report could be more detailed, the key metrics will represent the different concepts presented in the framework: profitability, solvency risks, and liquidity risks.

A current mockup of such a scorecard can be found below. It shows that, in this example, we are taking too much market risk compared to the budget (illustrative, set at the Surplus Buffer level).

This report currently doesn’t aggregate risks (mainly asset, market, and credit risk), treating them as fully uncorrelated. This is not accurate but not significant as currently, only market risk is significant and can be managed by allocating smaller targets for each risk.

Open questions to the community

• This framework adapts some traditional finance concepts to the reality of DeFi. Should the framework follow more directly an established TradFi standard like Basel III?
• How should the MKR minting be seen? Something dramatic that should happen only when everything goes wrong?.Or something that is within the scope of business as usual (i.e. once or twice per year). This will influence the size of the surplus buffer (for the same level of risk).
• Do you have a view on the constraints, of both solvency and liquidity, that MakerDAO should decide?

Conclusion

As we have seen, ALM can be seen as a constrained optimization problem where balance sheet items and the market environment are the inputs and should be arranged in order to optimize value-creation for MKR holders while checking the risk constraint that protects DAI holders.

A framework was proposed that led to the first draft of an ALM report.

Going forward, the following items need to be addressed:

• The risk constraints should be defined and approved by the MakerDAO community;
• More work needs to be done on a term structure of interest rate in DeFi and comparison to the TradFi one;
• DAI maturity profile should be studied;
• A more elaborated monthly ALM report should be presented to the community