[512m] A Quantitative Assessment of dTRINITY’s Subsidized Interest Rate Model

This article was published originally by 512m Research.

Introduction

In our previous article, we provided an in-depth examination of the mechanisms underlying dTRINITY’s new decentralized stablecoin, known as dUSD. While dUSD is a pivotal element within their ecosystem, it constitutes only one part of the DeFi Trinity. In this piece, we turn our attention to the second component: dLEND. This article adopts a more technical perspective on the lending and borrowing protocol, elucidating how dLEND as a platform compares against existing lending platforms like AAVE.

dLEND is a subsidized lending protocol on Fraxtal L2, designed to harness the yield that naturally accrues from the yield-bearing collateral of dUSD. This yield, in turn, is used to reduce the interest expenses typically incurred by borrowers. The resulting decrease in effective borrowing costs opens the door to creative leveraged strategies, such as looping, in which users repeatedly borrow against their assets to reinvest in the protocol.

A visualisation of dUSD’s borrowing rate model on dLEND, before and after subsidies.

We aim to systematically simulate, test, and compare the performance of a looping strategy using Ethena’s sUSDe (Staked USDe) as the base on dLEND against using AAVE standard and AAVE E-Mode.

A screenshot of app.dtrinity.org showing a lower Net Borrow APY for dUSD than its Supply APY.

Simulating the Markets

To achieve our objective, we center our analysis on what we refer to as the “Stablecoin Prime Rate” (SPR). This Index, which we developed, is primarily a TVL-weighted composite that reflects the Supply Annual Percentage Rate (APR) of the largest DeFi protocols. Historical data pertaining to this Index enables us to calibrate parameters to reflect its fluctuations over time, offering a robust foundation for our simulations. Below is a figure showing the development of the SPR over the past year:

Historical SPR Index over the past 365 days

In order to simulate the SPR, our approach employs a normal walk, incorporating both mean reversion and a random jump factor. This design choice is intended to capture real-world market dynamics, wherein significant macroeconomic events may cause abrupt spikes (either upwards or downwards) in interest rates. The base rate for our simulation is initially set at an SPR of 15 percent, and each scenario is ran across 500 distinct random walks. This comprehensive modelling allows us to observe and quantify how the looping strategy on dLEND performs under a multitude of hypothetical market conditions, from stable periods to sudden and severe fluctuations.

Below, we present a figure illustrating an excerpt of 25 out of the 500 random walks, offering a representative snapshot of the diverse trajectories the SPR may follow over time. This shows that we are simulating a wide variety of market conditions under the influence of sporadic jumps and mean reversion.

Selected 25 SPR simulation paths

In addition, we classify each of the simulations into one of three distinct market regimes: “bullish,” “bearish,” or “sideways,” based on the prevailing direction of the Index’s price movement. This categorization provides a critical framework for the analyses that follow, enabling a more granular understanding of the varying outcomes observed under different market conditions.

Subsequently, we proceed to simulate the yield for sUSDe. It is logical to assume that sUSDe’s Annual Percentage Yield (APY) is influenced by movements in the SPR, given that it reacts to a variety of market factors. However, relying solely on a simple linear regression to map this relationship would be unduly simplistic. In order to better capture the non-linear dynamics reflected in historical data, we fit multiple models to estimate the connection between the SPR and sUSDe. Among the models tested, random forest regressions and gradual boosting regressions emerged as the top two in terms of predictive accuracy. Their respective results are presented in the figure below.

Regression results for Random Forest and Gradient Boosting on SPR/sUSDe APY

Building upon the fitted gradient boosting regression model, the next step involves inputting the simulated SPR price movements to generate a corresponding sUSDe APY. In doing so, we further refine our approach by incorporating the market regime classification, thereby adjusting the simulated sUSDe APY to align more closely with observed historical trends. To introduce an additional layer of realism, we incorporate a small degree of stochastic noise into the modelling process.

The resulting simulated sUSDe APY values, derived from 25 sample SPR simulations, are presented in the figure below, to better show how our model’s projections for sUSDe APY evolve in response to different market conditions and random perturbations.

Selected 25 simulated sUSDe APY based on GBR

Next, we leverage the SPR to estimate utilization rates in dLEND and in both AAVE configurations. As one might expect, during periods characterized by heightened demand for stablecoins, indicated by an elevated SPR, utilization rates in lending protocols tend to rise accordingly. Thus, our modelling of utilization rates for each protocol is directly influenced by the SPR, albeit with certain distinctions across the different pools.

On dLEND and in AAVE’s E-Mode, higher Loan-to-Value (LTV) ratios enable more extensive leverage/looping, which frequently drives utilization rates to approach, and sometimes exceed, the protocols’ optimal thresholds. To simulate these effects, we employ a sigmoid function in conjunction with historically established minimum and maximum utilization rates, as well as a market sentiment adjustment based on the SPR. Additionally, a small stochastic component is introduced to capture the inherent variability in real-world market conditions. The results of the simulated utilization rates can be seen below.

Selected 25 simulated utilization rates for all 3 protocols based on the SPR

Building upon the simulated utilization rates for each protocol, we employ the respective Interest Rate Models (IRM) to derive the corresponding borrowing rates. The resulting simulated borrowing rates are illustrated in the figure below, and from these, we identify several key observations.

Selected 25 simulated borrow rates for all 3 protocols based on their utilisation rates

First, consistent with historical patterns and the assumptions embedded in our model, users on both dLEND and AAVE’s E-Mode exploit higher LTV ratios, thereby driving utilization rates to or beyond the optimal threshold. This phenomenon sometimes triggers sudden spikes in borrowing costs. In turn, these elevated costs tend to diminish demand for stablecoin borrowing and leverage, subsequently forcing utilization rates and borrowing rates to normalize. Although similar spikes appear in AAVE’s standard markets, they are generally less pronounced than in the high-LTV segments.

Another notable insight is that dLEND’s borrowing rate frequently drops below zero percent. This outcome reflects the impact of subsidized borrowing costs on dLEND, which are derived directly from the SPR.

Assessing the Strategies

Having now assembled the necessary inputs, we can proceed to analyze the looping strategies across each protocol. For the purposes of this simulation, we assume an initial capital of 10,000 USD, which is allocated to sUSDe and subsequently looped to the maximum feasible degree. For simplicity, we do not factor in transaction fees or liquidation events in this model. Should borrowing costs exceed the yield originating from the collateral, the shortfall is simply deducted from the overall position value.

The graph below illustrates the resulting trajectory of each strategy’s portfolio value. Notably, while all three strategies exhibit broadly similar behavior, dLEND and AAVE’s E-Mode demonstrate a higher potential maximum performance, reflecting the advantages conferred by their respective higher LTV ratios.

Equity backtest for selected 25 simulations on each protocol

To gain deeper insight into each strategy’s risk and return profile, we plot the distribution of final portfolio values after 365 days for all 500 simulations. The figure below visualizes these distributions, revealing notable distinctions in both average returns and variability.

As indicated, AAVE E-Mode achieves the highest mean final portfolio value, at approximately 16,660 USD. This is largely attributable to its higher maximum LTV ratio of 0.9, which permits up to 10 loops, greater than dLEND’s limit of 5 loops and AAVE Standard’s limit of 4. By comparison, dLEND attains a mean final value of around 16,248 USD, outperforming AAVE Standard’s approximate average of 13,526 USD.

Nevertheless, the improved mean performance of AAVE E-Mode is accompanied by a substantially broader dispersion. Its standard deviation of about 2,402 USD indicates a higher propensity for extreme outcomes, including occasional net losses. Conversely, both dLEND and AAVE Standard exhibit lower variability, reflecting more consistent performance but, on average, slightly lower returns than E-Mode.

Distribution of final portfolio values over 500 simulations

Whereas the preceding graph aggregates outcomes across all market conditions, we also sought to illustrate each strategy’s performance under distinct market regimes — namely bullish, sideways, and bearish. Utilizing the same classification methodology described earlier, we plot the final portfolio value for each strategy, stratified by these three regimes.

Based on the 500 simulations, AAVE E-Mode continues to exhibit the highest mean final value of around 17,900 USD in bullish markets. In sideways environments, however, the advantage of E-Mode diminishes considerably, and in bearish conditions it even falls short of dLEND, with a mean of approximately 14,280 USD compared to dLEND’s 14,613 USD. Across all three regimes, AAVE Standard tends to underperform relative to dLEND, reinforcing the latter’s competitiveness as a stable yet high-yielding alternative.

Distribution of final portfolio values over 500 simulations, grouped by market regime

Conclusion

In conclusion, our simulations indicate that both dLEND and AAVE E-Mode deliver compelling results for users seeking to implement leveraged looping strategies under a range of market conditions. AAVE E-Mode demonstrates a higher average return when markets trend bullish, owing to its elevated LTV threshold and capacity for additional loops. However, this advantage comes at the cost of greater volatility, as evidenced by the larger standard deviation and the occasional likelihood of net losses. Thus, while E-Mode can generate superior returns, it simultaneously introduces a more pronounced risk profile.

dLEND, by contrast, offers a more balanced proposition. Although its maximum returns typically fall just shy of E-Mode’s figures, dLEND’s outcomes are notably more predictable. This relative stability is primarily attributed to its yield-subsidisation mechanism, which often allows borrowing rates to venture below zero. As a result, individuals who prioritize consistency over aggressively optimizing returns may find dLEND’s risk-return profile particularly attractive, especially in sideways or bearish market conditions, where it has outperformed E-Mode in our simulations.

From a user’s perspective, selecting between dLEND and E-Mode should hinge on one’s tolerance for risk and expectations regarding future market conditions. Those who anticipate a bullish trend and are comfortable with higher volatility might opt for E-Mode, aiming to reap outsized gains from its superior looping capabilities. Conversely, users who expect prolonged periods of reduced volatility or bearish sentiment, or who prefer more dependable yields, may well find dLEND to be the more suitable choice.

Ultimately, our findings underscore the importance of tailoring leverage strategies to individual risk profiles and market outlooks. Each platform exhibits strengths and weaknesses that become more or less pronounced depending on underlying conditions. While E-Mode maximizes upside potential in favorable scenarios, dLEND delivers steadier performance across a broader range of market regimes. By carefully evaluating these trade-offs, professional loopers can better align their strategies with the dynamic realities of decentralized finance.

---

📢 Join the dTRINITY community to get the latest updates!

Website | X (Twitter) | Documentation | Discord | Blog | Other links

Disclaimer: dTRINITY is not available to residents of Canada, Iran, North Korea, Russia, the USA, the UK, and other restricted regions.

The information contained herein should not be considered legal, business, financial, or tax advice. Past performance is not indicative of future results. Digital assets and DeFi protocols carry significant risks, including the potential for complete loss of funds. By using dTRINITY, you acknowledge and accept these inherent risks. View our full Disclaimer and Terms to learn more about the risks involved.