DeFi Activity, Stablecoins, and the Impact on Price Volatility: Part 4

In Part 4 of this series, we move beyond correlation and establish that stablecoin lending activities, particularly repayment frequency, causally predict Ethereum (ETH) volatility using methods like Granger causality and ARX modeling. The analysis reveals that USDC repayments signal next-day volatility, USDT predicts over several days, and DAI reflects gradual, deliberate market shifts, offering traders and risk managers tailored, actionable insights to anticipate market turbulence more accurately.
In Part 1, Part 2 and Part 3, we provided a comprehensive foundation exploring how stablecoins - specifically USDC, USDT, and DAI - anchor the decentralized finance (DeFi) ecosystem by stabilizing market conditions through distinct collateralization methods. Initially, we detailed the operational mechanics of these stablecoins, highlighting the transparency of USDC’s fiat-backed model, the widespread yet less transparent adoption of USDT, and the decentralized, crypto-collateralized governance structure of DAI. Subsequently, we uncovered significant correlations between stablecoin lending activities - particularly repayment frequencies - and Ethereum (ETH) volatility, positioning repayment behaviors as reliable early indicators of market stress.
Our earlier analysis further identified "shock events," periods marked by extreme repayment activity closely aligned with real-world market disruptions, demonstrating their systemic nature rather than isolated incidents. These insights underscored the value of monitoring repayment activity to anticipate volatility spikes.
We move beyond correlation to rigorously assess causality. We now investigate whether past stablecoin lending activities genuinely forecast future Ethereum volatility changes. Leveraging advanced statistical methodologies - including Autocorrelation (ACF), Partial Autocorrelation (PACF), Granger causality tests, and enhanced predictive modeling - we aim to determine if stablecoin lending not only coincides with volatility but actively drives it. By establishing clear predictive relationships, our analysis provides traders, investors, and risk managers with precise, actionable insights into the mechanisms underlying market volatility, enabling improved strategic decision-making in the rapidly evolving DeFi landscape.
For more detailed research and analysis, visit our research blog. Explore actionable market insights and stay ahead in the fast-changing DeFi landscape by connecting with Amberdata today.
Autocorrelation (ACF) and Partial Autocorrelation (PACF): How Long Does Past Volatility Influence ETH?
Autocorrelation (ACF) and Partial Autocorrelation (PACF) analyses are key methods in time series analysis that help identify how past data points influence current values. ACF measures the correlation between today's observations and their past values across different time lags, revealing how long historical patterns persist. PACF isolates direct correlations between observations, separated by specific time intervals, removing any indirect effects from intervening periods. Together, these tools determine the best lag structure for forecasting models, ensuring accurate capture of temporal dependencies. Applied to Ethereum volatility in the context of USDC, USDT, and DAI, ACF and PACF analyses clarify how past volatility influences current market conditions. For instance, a strong ACF at a one-day lag means today's volatility is heavily influenced by yesterday’s conditions. Likewise, a significant PACF value at a three-day lag indicates that volatility exactly three days prior directly impacts today's volatility, independent of the intermediate days.
USDC Volatility
Ethereum volatility linked to USDC tells an intriguing story of market responsiveness. Through Autocorrelation (ACF) and Partial Autocorrelation (PACF) analyses, we discovered a strong, immediate connection between today's volatility and recent market movements. Specifically, volatility showed pronounced autocorrelation up to seven days, indicating that what happens in the market today reverberates for about a week. The PACF emphasized this even more clearly, with a sharp spike at the very first lag - yesterday’s volatility - highlighting an immediate market reaction.
Why is this particularly meaningful for USDC? As a stablecoin known for transparency and fiat-backing, USDC attracts institutional traders who actively manage liquidity. When market sentiment shifts even slightly, these traders swiftly rebalance portfolios, directly affecting volatility the next day. For those watching closely, this means monitoring short-term repayment activities and volatility patterns in the USDC space can be a powerful forecasting tool for anticipating rapid market shifts.
USDT Volatility
The volatility dynamics around USDT reflect a slightly different narrative, shaped by its widespread usage among both institutional and retail users. ACF and PACF plots reveal volatility that is strongly tied to recent history, particularly at lag one, but this connection gradually weakens over a week-long period. Unlike USDC’s sharp and immediate reaction, USDT’s PACF indicates a slower buildup of volatility influence over several days.
This gradual accumulation makes sense when considering the diverse nature of USDT's user base. Institutions and retail traders worldwide react differently, and at varying speeds, to market signals. Consequently, volatility predictions for USDT aren’t about instantaneous reactions but rather about understanding how repayment activities accumulate and signal broader market sentiment shifts over multiple days. For traders, interpreting this pattern means keeping a close eye on repayment frequencies that gradually escalate, signaling emerging market instability before it fully materializes.
DAI Volatility
DAI volatility offers yet another distinct perspective, shaped by its decentralized, crypto-collateralized structure. Through ACF analysis, we observed persistent autocorrelation up to a week, suggesting a steady market memory. The PACF revealed strong, immediate effects at lag one that diminish thereafter, underscoring the importance of recent market conditions but highlighting a distinctly cautious approach compared to USDC and USDT.
The slower, deliberate adjustments observed in DAI's volatility reflect the decentralized and community-driven nature of its users. Decisions in the DAI ecosystem involve careful risk assessments and incremental position adjustments. This means volatility signals appear more gradually but carry a sustained impact. Traders focusing on DAI can gain valuable predictive insights by paying attention to these cumulative repayment behaviors unfolding over extended periods. Understanding this cautious approach allows market participants to better anticipate shifts and position strategically in advance of emerging volatility.
Granger Causality: Do Stablecoin Activities Predict ETH Volatility?
Having established the temporal connections through ACF and PACF analyses, the next critical question becomes whether stablecoin lending activities genuinely drive Ethereum volatility - or simply move alongside it. To answer this definitively, we turn to Granger causality testing, a statistical method specifically designed to identify directional predictive relationships. In simpler terms, Granger causality helps us determine if changes in stablecoin activity can reliably predict future shifts in volatility rather than just coinciding with them.
USDC Causality
Our tests for USDC revealed a strikingly clear predictive relationship, particularly from repayment metrics. Repayment frequency, measured by the number of repayments (numberOfRepays), demonstrated extremely significant causality with ETH volatility at a one-day lag (p-value ≈ 1.26e-12). This immediate causality aligns naturally with USDC’s institutional-heavy usage: institutions quickly rebalance their holdings through repayments in anticipation of imminent volatility. Withdrawals also showed strong predictive power, underscoring rapid liquidity shifts as early warnings of volatility. Thus, USDC repayments and withdrawals serve as powerful leading indicators of next-day ETH volatility, directly reflecting swift institutional risk management strategies.
USDT Causality
In contrast, the Granger causality results for USDT told a more gradual story. While repayment frequency remained the strongest indicator, its predictive strength peaked noticeably later, around a seven-day lag (p-value ≈ 3.4e-05). This slower buildup highlights the diverse, global user base of USDT, including retail traders whose responses are more measured and unfold gradually. Borrowed amounts (borrowedUSD) also emerged as predictive at intermediate lags, suggesting incremental leverage accumulation that gradually influences volatility. The findings emphasize that understanding USDT-related volatility requires patience, with market signals accumulating steadily before clearly indicating upcoming turbulence.
DAI Causality
DAI’s Granger causality results reinforced its unique decentralized and cautious character. Like USDT, repayment frequency was the dominant predictive factor, reaching peak significance at a seven-day lag (p-value ≈ 3.49e-08). However, unlike the rapid responses seen with USDC, DAI’s delayed predictive signals highlight careful and incremental position adjustments by its community-driven user base. The market does not react impulsively but rather builds deliberate repayment patterns over several days, making the predictive signals highly robust yet slower to emerge. This reflects DAI's decentralized governance model, where volatility indicators become clear only after deliberate community actions accumulate.
Across all three stablecoins, repayment frequency consistently emerged as the most potent predictor of Ethereum volatility. Yet, crucial differences in timing - immediate for USDC, gradual for USDT, and deliberate for DAI - reflect their unique market roles and user behaviors. Traders and risk managers leveraging these nuanced insights can better anticipate volatility spikes, strategically positioning themselves ahead of market turbulence.
Baseline AR Models: Establishing a Predictive Benchmark
Earlier, our Granger causality tests showed clearly that stablecoin lending activities predict Ethereum (ETH) volatility. Before incorporating these external market signals into our models, we first wanted to establish a baseline - essentially asking: how well can past volatility alone predict future volatility? To answer this, we used Autoregressive (AR) models, which forecast today's volatility based solely on historical volatility data.
In intuitive terms, an AR model assumes:
Today's Volatility = (Influence of Past Volatility) + (Random Noise)
Expressed mathematically, the AR formula looks like this:
Here’s what each part means intuitively:
- Volatilityₜ: Today's Ethereum volatility that we're aiming to predict.
- βᵢ (Volatilityₜ₋ᵢ): How strongly volatility from previous days (up to a week earlier) influences today's volatility.
- εₜ: Random fluctuations or unpredictable market noise not explained by historical data alone.
By examining these relationships, we establish how persistent Ethereum volatility truly is - meaning how much today's market movements depend on recent history. This baseline evaluation helps us answer key questions:
- Does volatility consistently cluster in time (autocorrelation)?
- How accurately can past volatility alone predict future volatility?
- Is there significant room to enhance predictions by incorporating external indicators, like stablecoin lending activities?
Setting this baseline clearly shows how much predictive power stablecoin metrics add when we integrate them later, highlighting their true value in anticipating market shifts.
USDC Baseline AR Model
For USDC, our baseline Autoregressive (AR) model offered modest explanatory power, capturing about 10% of the variability in Ethereum (ETH) volatility (R² = 0.100). This relatively limited predictive strength indicates that historical volatility alone provides only a basic understanding of future market movements, leaving considerable room for improvement.
Digging deeper, we observed significant volatility persistence at specific time intervals: volatility today was notably influenced by very recent conditions - specifically one and two days prior - as well as somewhat older patterns from six days earlier. These lags suggest that volatility exhibits both immediate and cyclical persistence, reflecting institutional trading behaviors associated with USDC’s transparency and fiat-backed stability.
In terms of predictive accuracy, the baseline model achieved a Mean Absolute Error (MAE) of 0.0112 and a Root Mean Squared Error (RMSE) of 0.0167. While these figures establish a useful benchmark, they highlight substantial forecasting limitations when relying solely on historical volatility.
These findings clearly illustrate the necessity of incorporating external stablecoin lending metrics, such as repayment frequency, to better anticipate volatility shifts. Ultimately, this baseline establishes a foundation for evaluating the predictive enhancements introduced by including additional market indicators.
USDT Baseline AR Model
The baseline AR model for USDT showed notably stronger explanatory power compared to USDC, explaining roughly 36% of Ethereum volatility fluctuations (R² = 0.358). This higher value indicates stronger persistence in volatility within the USDT ecosystem, reflecting its broad adoption and diverse user base spanning institutional and retail investors.
Examining the model closely, we found significant volatility persistence at lags 1, 2, and 6 days. This pattern suggests volatility today heavily reflects both recent and slightly older market conditions, highlighting sustained reactions from the global, mixed-market participants who utilize USDT extensively.
Despite improved explanatory power, the model’s predictive accuracy remained similar to USDC, with a Mean Absolute Error (MAE) of 0.0114 and a Root Mean Squared Error (RMSE) of 0.0171. These performance metrics emphasize inherent limitations in predicting future volatility solely based on past market behavior, reinforcing the potential benefit of integrating external lending activity metrics - such as repayment frequency - to strengthen volatility forecasts in the USDT context.
DAI Baseline AR Model
For DAI, our baseline AR model captured approximately 30% of Ethereum volatility variations (R² = 0.305), placing it between USDC’s modest explanatory power and USDT’s more substantial persistence. This intermediate result aligns well with DAI’s decentralized governance and cautious user behavior, leading to measured, incremental market responses.
We observed significant volatility persistence at multiple intervals - lags 1, 2, 4, and 6 - indicating volatility is shaped by both immediate past market conditions and periodic historical patterns. This intermittent influence underscores the deliberate decision-making processes typical in the DAI ecosystem, where market participants slowly react to changing conditions rather than making swift adjustments.
Model accuracy was lower than for USDC and USDT, reflected by a higher Mean Absolute Error (MAE) of 0.0131 and Root Mean Squared Error (RMSE) of 0.0208. These elevated errors imply that historical volatility alone struggles to fully account for DAI’s volatility behavior, leaving notable gaps in predictive capability. Thus, incorporating external stablecoin activity metrics - particularly repayment frequencies - appears essential for capturing the complexities of DAI volatility and improving forecasting accuracy.
Extended ARX Models: Adding Stablecoin Activities to Improve Predictions
Our baseline AR models clearly demonstrated that Ethereum volatility is influenced by its own past behavior. However, those models didn't account for broader market activities, such as stablecoin lending behaviors, that may also predict volatility. To address this, we introduced Extended Autoregressive models with eXogenous variables (ARX), which combine historical volatility data with external market indicators to enhance forecasting.
In simple terms, an ARX model assumes:
Today’s Volatility = (Influence of Past Volatility) + (Influence of Stablecoin Activities) + (Random Noise)
Expressed mathematically, the full ARX formula is:
Here’s what each component means intuitively:
- Volatilityₜ: Today's Ethereum volatility we're trying to predict.
- βᵢ (Volatilityₜ₋ᵢ): How past days’ volatility (up to 7 days ago) impacts today's volatility.
- γⱼ (Xₜ₋ⱼ): How stablecoin-related activities from previous days influence today’s volatility.
- εₜ: Represents randomness or unpredictable market fluctuations.
The stablecoin activity metrics selected as external variables (X) include:
- USDC: number of repayments (numberOfRepays), withdrawal amounts (withdrawnUSD), and withdrawal frequency (numberOfWithdraws)
- USDT: number of repayments (numberOfRepays), repayment amounts (repaidUSD), and borrowed amounts (borrowedUSD)
- DAI: number of repayments (numberOfRepays), number of withdrawals (numberOfWithdraws), and number of borrowings (numberOfBorrows)
These metrics were carefully chosen based on previous findings demonstrating their strong predictive relationships with volatility, especially repayment frequency, which has consistently emerged as a reliable indicator. By incorporating stablecoin activities into our volatility forecasts, ARX models provide deeper market insights and significantly improve predictive accuracy beyond historical volatility alone.
USDC ARX Model
Integrating stablecoin lending activities significantly enhanced our model’s predictive strength for USDC. Specifically, the explanatory power increased from an R² of 0.100 (baseline AR model) to 0.176. Among the lending metrics, repayment frequency (numberOfRepays) from one and two days earlier emerged as especially significant predictors (p < 0.05). These findings align with our earlier analyses highlighting the immediate impact of institutional traders who rapidly adjust positions in anticipation of market shifts. While the Root Mean Squared Error (RMSE) slightly increased from 0.0167 to 0.0188 due to added complexity, the improved model accuracy strongly indicates that repayments serve as valuable early indicators, clearly reflecting proactive institutional responses to expected volatility.
USDT ARX Model
For USDT, the inclusion of stablecoin lending behaviors substantially boosted our model's explanatory ability, with the R² increasing from 0.358 to 0.439. This notable improvement underscores how repayment frequency (numberOfRepays) over several previous days significantly influences volatility forecasts. The cumulative predictive strength of repayment activity confirms our earlier observation: the diverse, global user base of USDT, including both retail and institutional traders, responds gradually rather than instantaneously to market signals. Despite increased complexity, this integration considerably enhances predictive accuracy, equipping traders and risk managers with clearer foresight into future volatility trends.
DAI ARX Model
Incorporating lending activity metrics similarly improved volatility forecasting for DAI, raising the model’s explanatory power from an R² of 0.305 to 0.387. Once again, repayment frequency (numberOfRepays) across multiple lags proved especially valuable, mirroring earlier insights into DAI's deliberate and incremental community-driven decision-making process. Although the RMSE showed a minor increase (from 0.0208 to 0.0212), the significant enhancement in predictive capability strongly justifies this complexity. DAI’s decentralized structure and cautious repayment behaviors effectively signal future market movements, allowing traders to anticipate volatility shifts more accurately.
Overall, integrating stablecoin lending activities - particularly repayment frequencies - significantly strengthens Ethereum volatility forecasts. Beyond merely reflecting historical volatility, ARX models offer deeper market insights by capturing proactive market responses. This additional predictive accuracy provides traders and risk managers a meaningful advantage in navigating upcoming volatility events within USDC, USDT, and DAI ecosystems.
Conclusion
Our comprehensive analysis establishes that stablecoin lending activities - particularly repayment behaviors - serve as robust predictors of Ethereum volatility, though each stablecoin ecosystem offers unique insights. With USDC, characterized by institutional dominance and transparent fiat backing, repayments act as immediate volatility signals. Institutional traders swiftly adjust positions, making repayment activity a valuable next-day volatility indicator. In contrast, USDT’s mixed institutional and retail user base results in more gradual responses. Repayment activities accumulate slowly over several days, signaling upcoming volatility with delayed yet sustained predictability. This highlights the importance of monitoring cumulative repayment patterns within the globally diverse USDT ecosystem. For DAI, governed by decentralized decision-making and cautious community actions, repayment behaviors signal volatility shifts methodically. Market reactions emerge slowly, providing clear, persistent signals after about one week.
Practically, our ARX models confirmed that integrating stablecoin metrics significantly enhances volatility forecasting. USDC lending data substantially improved model accuracy, underscoring quick institutional market responses. For USDT, repayment frequencies added notable predictive depth by capturing incremental volatility shifts. Similarly, DAI's decentralized lending patterns strengthened predictions, emphasizing deliberate community-driven market adjustments.
These findings suggest traders and risk managers should tailor volatility strategies by the stablecoin ecosystem, closely monitoring immediate repayment signals for USDC, cumulative behaviors for USDT, and incremental shifts for DAI. Ultimately, leveraging these nuanced insights allows market participants to anticipate Ethereum volatility more effectively, providing a strategic advantage in navigating the evolving DeFi landscape.
Read more expert commentary and market research on our research blog. Explore how Amberdata’s powerful analytics tools can strengthen your DeFi decisions. Contact us to get started, or request a demo for a personalized experience.
Disclaimers
The information contained in this report is provided by Amberdata solely for educational and informational purposes. The contents of this report should not be construed as financial, investment, legal, tax, or any other form of professional advice. Amberdata does not provide personalized recommendations; any opinions or suggestions expressed in this report are for general informational purposes only.
Although Amberdata has made every effort to ensure the accuracy and completeness of the information provided, it cannot be held responsible for any errors, omissions, inaccuracies, or outdated information. Market conditions, regulations, and laws are subject to change, and readers should perform their own research and consult with a qualified professional before making any financial decisions or taking any actions based on the information provided in this report.
Past performance is not indicative of future results, and any investments discussed or mentioned in this report may not be suitable for all individuals or circumstances. Investing involves risks, and the value of investments can go up or down. Amberdata disclaims any liability for any loss or damage that may arise from the use of, or reliance on, the information contained in this report.
By accessing and using the information provided in this report, you agree to indemnify and hold harmless Amberdata, its affiliates, and their respective officers, directors, employees, and agents from and against any and all claims, losses, liabilities, damages, or expenses (including reasonable attorney’s fees) arising from your use of or reliance on the information contained herein.
Copyright © 2025 Amberdata. All rights reserved.
Michael Marshall
Mike Marshall is Head of Research at Amberdata. He leads pioneering research initiatives at the forefront of blockchain and cryptocurrency analytics. Mike is a seasoned quantitative analyst with a 15-year track record in developing AI-driven trading algorithms and pioneering proprietary cryptocurrency strategies. His...