The Target Rate Feedback Mechanism: An Introduction

Background

The Target Rate Feedback Mechanism (TRFM) has been mentioned a few times in recent memory, for example during governance calls or in forum posts like this one. However, the TRFM is not widely understood, so I’m making this post to provide the community with an easy introduction to the concept. There’s no intent here to suggest implementing this mechanism; rather, I merely want to clearly explain what the TRFM is so that everyone can understand and participate in any discussions into which it enters.

The TRFM can be thought of (roughly) as an alternative to the DAI Savings Rate (DSR) that has more flexibility in terms of applying directional pressure to the DAI price. It has been discussed since the early days of the Maker project, although work on it was stopped some time ago. This steemit post describes the mechanism and some of the early research that was done, and is still a good reference. A recent Medium article describes the same idea and is also recommended reading for those wanting to dig deeper. Here, I’ll explain only the most basic ideas needed to understand the TRFM.

Definitions

Reference Asset

An asset the system uses to measure value. This is currently the United States Dollar (USD). Oracle prices must be reported in units of this asset.

Market Price (MP)

The value, according to the market, of one DAI against the reference asset. This is measured through, for example, DEX trades, and can be seen on sites like https://dai.stablecoin.science/.

Target Price (TP)

This is the value, in the reference asset, that the system targets for one DAI. I.e., if the system is in equilibrium, the target price and market price are equal. This value is also called the “redemption price”. It affects how much DAI can be drawn against a fixed value of collateral, and also how much collateral is received per-DAI during Emergency Shutdown. The target price has been fixed at $1 since the system launched. It is stored in the par variable of the Spotter smart contract. An example: at a 150% collateralization ratio and a $1 target price, 100 DAI can be drawn against $150 of ETH collateral. If the target price is changed to $2, only 50 DAI can be drawn against the same $150 of ETH.

Target Rate (TR)

The rate at which the target price is changing, usually thought of and expressed as a yearly percentage. Currently, the Target Rate is zero and the system does not track a value for it.

Target Rate Feedback Mechanism (TRFM)

A feedback mechanism that adjusts the target rate in order to equalize the target price and the market price. The rest of this post will discuss this in more detail, but the diagram below gives a basic illustration.

TRFM

Implications of the TRFM

The TRFM involves a major conceptual shift. The entire world thinks of DAI as being a dollar-pegged stablecoin, but if the TRFM is implemented, then DAI no longer has a fixed peg. It still has a peg (the target price) at any given time, but this peg changes based on the target rate. DAI with a TRFM could instead be thought of as a “volatility smoothed” or “semi-stable” coin (the term “reflex bond” has recently come to be associated with such an asset). Here’s a hypothetical example of how this could play out in practice: suppose DAI initially has a TP and MP of $1, and a TR of 0%/year. Then, an extreme market event causes the MP to drop to $0.90. The TRFM activates and sets the target rate to 10%/year to incentivize supply contraction and demand to hold DAI. Suppose that after 8 months, the TP and MP are equal again, so the TR returns to zero; now the target price (and also the market price, since we said they were the same) will be ~$1.0656.

Another important feature of the TRFM is that it can be negative as well as positive. This means that it can deal equally well with DAI prices that are above and below the target price. This contrasts with the present situation in which the DSR is at 0% and cannot be lowered further, despite DAI remaining above peg. In fact, if DAI had a TRFM, having the DSR as well would be of questionable value (and very confusing for users). Further, stability fees could remain at non-zero values regardless of the sign of the target rate, ensuring that the system can still earn risk-premium income while the market overall is driven to resolve the discrepancy between the target and market prices.

TRFM Incentives

The ways in which the TRFM creates incentives to equalize the target and market prices are somewhat subtle and worth examining in more detail.

Market Price < Target Price

If DAI is trading below its target, the target rate will be set to a positive value. This implies an increasing target price.

Vault user viewpoint

As the target price increases, Vault collateralization levels decrease (roughly, the system is viewing DAI as being “worth more” and hence each Vault’s dollar-valued debt is increasing). This incentivizes Vault holders to either buy DAI to repay debt (a good deal, since DAI is trading low) or to add collateral. Only the former will impact the price of DAI and thus the target rate (which a Vault holder would like to see reduced), hence at a high enough target rate this becomes the preferred option and creates demand for DAI, driving up the price.

DAI holder viewpoint

The rising target price represents and effective savings rate on DAI, incentivizing the holding and acquisition of DAI. Generally, this comes about because the price Vault holders are willing to pay will increase over time (to avoid loss of collateral), and because one DAI will be able to claim a greater quantity of collateral should Emergency Shutdown occur.

Market Price > Target Price

If DAI is trading above its target, the target rate will be set to a negative value. This implies a decreasing target price.

Vault user viewpoint

As the target price decreases, Vault collateralization levels increase (i.e. the same amount of DAI debt is regarded as a lesser dollar amount by the system). This frees up Vault holders to mint more DAI without increasing their liquidation risk, incentivizing a DAI supply increase, which should lower the market price. This is functionally similar to a negative stability fee, but note that the system can still collect non-zero fees (albeit in an inflating currency) even with a negative target rate.

DAI holder viewpoint

Since Vault holders will be willing to accept progressively lower maximum prices for DAI in the future as the target price decreases, DAI holders are incentivized to sell their DAI to lock in the benefits of the current high market price, and those thinking of holding DAI as a store of value are likely to reconsider. There is also the fact that the amount of collateral DAI will be able to claim during ES is decreasing, weakening the potential payoff of holding DAI if ES occurs.

Implementing the TRFM

Note: this section is intended to address curiosities around how the TRFM might be implemented, and does not represent a proposal for actually doing so–that would require a proper technical design document presented as part of a MIP.

At a minimum, three new components would be required to fully implement an automated TRFM in the present DAI Credit System:

  • a DAIUSD price oracle
  • a target price calculator
  • a target rate controller

The DAI price oracle could be spun up using the existing oracle infrastructure; the target price calculator (which takes the target rate as an input) would be a trivial contract similar to the Pot or Jug (in fact, simpler than either of them). The target rate controller is more involved, as it must specify an algorithm for calculating the target rate.

The options for implementing the algorithm that controls the target rate are quite nearly limitless. However, a PID controller is a simple and widely-used process control algorithm that has been explored previously and could be implemented fairly easily on-chain. The purple paper even sets a naming precedent for some of the parameters that would be involved.

Alternatively, governance could set the target rate or even the target price directly, reducing the need for smart contract implementations. The downside of this is of course the introduction of the delays and caprices of human decision-making into the process, but something along these lines could serve as a temporary bridge to a fully-automated TRFM (which would likely need some tuning of parameters, at least initially). A trustless, unalterable (or at least predictably alterable) TRFM would be a much more reliable guide to market participants, as it eliminates uncertainty from the choices they must make.

15 Likes

great explanation Kurt.

I would like to see TRFM voted an implemented in the protocol. We may need it in the near future.

3 Likes

It’s certainly better than adding new collateral hoping that the DAI price will somehow magically return to $1.

I’d say this is the most important project for Maker since the introduction of adding new collateral types. Given our current monetary levers seem to be insufficient at keeping the arbitrary USD $1 peg, the TRFM would be an ideal fix. There will be some slight education needed for end-users to understand that we are no longer pegged to the $1 USD, but given our current situation I think people will find this to be a proactive step towards a more dynamic stable currency. There are also competitors seeking to implement reflexive bonds that could hurt Dai adoption.

So +1 to supporting this initiative. What would be the next step to create the MIP? Sufficient community support? I hope others can post to express the importance of this project so we can get started on a MIP.

1 Like

Can we use their approach? I think they want to use reflex bonds as collateral.

In principle I like the idea of building some adaptive system that is less reliant in human governance to keep the peg.

I find the idea fascinating. Some questions:

Say TP goes to 0.25 due to massive demand in dai. How does it translate into actual collateralization?

As TP decreases, I understand 1 dai could have less than 1usd worth behind eventually?

Personally, I always liked the idea of DeFi being grounded on overcollateralized money, as a departure from the massively undercolateralized fiat system. This will require a new mindset.

I guess dai holders will vote with their feet, but I worry about situations where things go out of control, and the system becomes very fragile.

Or is there some limit?

Say TP goes to 0.25 due to massive demand in dai. How does it translate into actual collateralization?

This is a very vague question, so I’ll try to answer for a couple different interpretations. In a realistic scenario, the target price would change gradually over time based on the target rate, and the collateralization levels would depend on the actions of Vault holders, etc, over that time period. If one were to artificially (say, by a gov action) just change the TP from $1.00 to $0.25, then collateralization level would increase by 4x (same amount of collateral, but DAI is suddenly worth 4 times less according to the system).

As TP decreases, I understand 1 dai could have less than 1usd worth behind eventually?

I’m not entirely sure how to interpret this question either, but: with a TRFM implemented, DAI has no fixed peg–it could become worth $0.72, $1.29, or really any other value depending on market forces.

DAI is still overcollateralized even if a TRFM is implemented (or, at least, it is no more at risk of undercollateralization with the TRFM than without it).

Or is there some limit?

Limit on what? This is all hypothetical right now; in general automated algorithms can and often do have operational bounds put in place to prevent extreme behavior. A TRFM could easily have built-in limits on e.g. the magnitude of the target rate or how often it was allowed to change. Again, this post isn’t an engineering design, it’s just an introduction to a concept.

2 Likes

I also think that once TRFM is in place we could start researching around different models/algos of it which can leads us to a whole new world of mechanism to achieve stability.
Stability is largely a manufactured product, not just 1:1 redemption

1 Like

Is the implication here if we added TRFM the DSR would end up on the deprecation path?

One of my first thoughts when reading this is “what happens to DAI locked in the DSR”, but I can save my specific questions concerning that for now.

I do not think this approximation is correct. The TRFM will in all likelyhood be used to offset US inflation only, with the DSR coming on top of this. It is correct that in the present low inflation situation we have had the last decades the DSR is likely to be low, but the two mechanisms still have different function IMHO.

1 Like

Yes, I see them as two separate instruments too, something like…
DSR -> T-bill
TRFM -> Continuos Incentive/disincentive Bond

Thank you for your answers. I want to make it clear I think this type of solution is probably where things will end up going. I’m only trying to understand the mechanics and implications.

Let me try to make my questions more specific.

In the current world, TP = 1usd, and LR is at 150% of Dai minted.
As a Dai holder, I know two things:

  1. There should be at least 1,5 dollars’ worth behind every dai, as enforced by the LR.
  2. During ES, I should be entitled to claim 1 usd worth of collateral.

In the TRFM world, if TP goes to 0,25, I want to understand how each of those two things change.
I understand during ES I’d only be entitled to claim 0,25 worth of colateral.
My question was, how does this impact the liquidation rate, and therefore the min colateralization to expect behind a Dai?

I hope this is less vague. Thanks!

Yes, that makes more sense, Alfablok:

With the TP at $0.25, then you are correct, during ES you can only claim $0.25 worth of collateral (according to the final oracle prices, and assuming the system is not undercollateralized).

The liquidation ratio is a separate parameter from the TP. Even if the TP has gone to $0.25, if the liquidation ratio is set at 150% then there’s still at least $0.375 (150%) value backing every DAI (at least under healthy operation of the system)

The goal of the TRFM is to set an intrinsic interest rate on the DAI price that maintains MP/TP equality. Trying to set a DSR value on top of this will upset that balance. If the target rate is positive, that is no difference from a positive DSR from the perspective of a DAI holder (their buying power is increasing). The big difference is that the target rate can go negative, whereas DSR cannot, allowing the TRFM to respond to any MP - TP deviation (whereas DSR, as we have seen, becomes useless under certain conditions).

So I don’t at all agree that TRFM and DSR are independent concepts that can trivially coexist. The have overlapping function and characteristics. Yes, they are not the same, but they are not totally distinct, either.

Yes, they are not independent since they are affecting Dai issuance, although I tend to see them affecting different scopes of incentives and time.

TRFM - Short time incentive/disincentive
DSR - Long time incentive.

This seems like a pretty BIG tradeoff to me.

I know that the peg is currently in a state that makes this more of an interesting concept, but in more stable economic environments I see DSR as a valuable tool (and key competitive advantage) to drive DAI adoption for more risk-off investors seeking yield. In the brief history that it has existed, we’ve seen very strong adoption so I would be very wary of ideas that abandon it in concept so early in its life.

I’m sure that the argument can be made that TRFM is theoretically able to also drive yield for DAI holders, but it appears to me that it also exposes the DAI holder to risks that are difficult to model and reason about which seemingly will hurt adoption.

4 Likes

The dollar peg is the PMF for Maker.

I’d much rather see Maker adopt open market operations than give up on the USD peg. See:

3 Likes

Why is the TRFM preferred to a negative DSR? They both achieve the same effect but one breaks DAI’s biggest selling point (parity with USD), while the other does not.

One reason to prefer the TRFM is ease-of-implementation: making DSR mandatory would require replacing the DAI ERC-20 contract which would be a major disruption to the ecosystem (probably requires an Emergency Shutdown followed by a full re-deployment, or at least that’s the cleanest way to do it). A TRFM also pays the negative rate to Vault holders by directly increasing their collateralization ratio–in the current system, a negative DSR would instead go to pay MKR holders.

Further, I think it’s a little disingenuous to claim that a negative DSR somehow preserves “DAI dollar parity” as if that’s a big advantage, because what they both do is decrease the purchasing power, over time, of holders of DAI, in a way that’s largely the same. And the purchasing power, I think, is what people really care about.

A concrete example: suppose you start with 100 DAI, which is at a $1 peg initially. Now consider two scenarios:
Negative 10% TRFM for a year: you still have 100 DAI, but it’s worth $90
Negative 10% DSR for a year: you have 90 DAI, worth $90

The loss of purchasing power is identical. Further, in the TRFM case, you could construct a chai-like wrapper token around DAI such that the balance fluctuates but the target value of a single token remains at $1.

2 Likes

Funding rates are not a novel invention, we should look to other perpetual synthetics for guidance.

You mention that TRFM compensates vault holders directly, while DSR does not. This is an invalid point because you can set negative SC to pay the negative DSR to vault holders.

I never said that I thought a negative DSR would preserve purchasing power. My argument was around price parity as DAI is a unit of account in my mind.

Of your remaining points, is it fair to say:

  1. TRFM is a lot more plausible to implement because we can avoid a shut down.
  2. If negative DSR can be implemented without a shut-down, it would be preferable

I hope to use DAI as a highly liquid USD quote currency one day. If we are going to price things in DAI, we must maintain 1USD = 1DAI. I think introducing a 2nd token creates confusion, reduces the brand and adds complexity. Only something we should consider if the alternative is indeed another difficult migration to new system.