A data-driven look at the connection between stablecoin lending trends and Ethereum volatility.
In our previous report, we explored how stablecoins like USDC, USDT, and DAI serve as critical anchors in the DeFi ecosystem. We did this by examining their underlying mechanisms and roles in maintaining market stability. Building on that foundation, Part 2 shifts our focus to a deeper analysis of Ethereum’s price volatility and its connection to lending activities within DeFi.
To capture a more comprehensive picture of market fluctuations, we now employ the Garman-Klass estimator, which leverages intraday price data (open, high, low, and close) rather than relying solely on closing prices. This refined volatility measure reveals the full range of daily price movements, offering insights that traditional methods might miss.
Additionally, we recognize that many lending metrics display trends over time (non-stationarity), potentially distorting our analysis. To address this, we applied the Augmented Dickey-Fuller (ADF) test to identify non-stationary variables and transformed them using first differencing. This ensures that our correlation analysis between lending behaviors and ETH volatility is based on stable, reliable data.
In this report, we detail how we merge ETH price data with lending metrics, calculate volatility using the GK estimator, and adjust for non-stationarity. The resulting insights offer traders and risk managers a clearer, data-driven perspective on how changes in lending activity may serve as early indicators of shifts in Ethereum’s market dynamics. For actionable market data and further insights into the evolving landscape of DeFi, connect with Amberdata today.
In this section, we introduce the Garman-Klass estimator—a refined approach to measuring volatility that leverages the full range of intraday price data. Unlike traditional volatility measures that rely solely on closing prices, the Garman-Klass method incorporates open, high, low, and close (OHLC) prices, providing a richer, more comprehensive view of market fluctuations.
Standard close-to-close volatility metrics can overlook significant price movements that occur within the trading day. When only the open and close prices are considered, important intraday dynamics—the peaks and troughs that occur as the market reacts to news or shifts in sentiment—are ignored. The Garman-Klass estimator addresses this gap by capturing both the overall price shift from open to close and the range between the highest and lowest prices observed during the period. This approach results in a more robust measure that better reflects the true market volatility.
Intuitively, consider the daily price movement of an asset as a journey. Measuring only the distance between the start and finish of this journey (open to close) provides only a partial view. It ignores the detours, sudden stops, and rapid changes in direction—the high and low points—that occur along the way. By incorporating all four key price points, the Garman-Klass estimator tells a fuller story of how much the price actually moved throughout the day.
The formula for the Garman-Klass volatility is as follows:
In this formula, we first use the natural logarithm of the ratio between the highest and lowest prices to capture the overall range of price movements during the period. This tells us, in proportional terms, how much the price fluctuated from its peak to its trough. Then, we also calculate the natural logarithm of the ratio between the closing and opening prices to measure the overall price change from the beginning to the end of the trading session.
The equation takes half of the square of the first logarithmic ratio to quantify the full extent of the intraday price range. In other words, this term shows us how wide the price swings were during the day. The second term adjusts this measure by accounting for the difference between the opening and closing prices; it multiplies the square of the second logarithmic ratio by a constant (derived from two times the natural logarithm of 2 minus one). This adjustment ensures that the net movement from open to close is properly reflected.
Together, these two parts give us a balanced picture of volatility. They combine to show not only the extreme fluctuations within the day but also the overall directional movement of the market during the session.
For our analysis, we applied the Garman-Klass estimator to each ETH OHLCV dataset, including ETH/USDC, ETH/USDT, and ETH/DAI. We compute the volatility so we can directly correlate these refined volatility estimates with our stablecoin lending metrics, helping to identify relationships between market behavior and DeFi activities.
Stablecoin lending and borrowing metrics often exhibit trends and shifts over time, so it is crucial to work with stationary data before analyzing their relationship with Ethereum’s volatility. To ensure our data was reliable, we applied the Augmented Dickey-Fuller (ADF) test to each numeric variable in our dataset—including metrics such as depositedUSD, borrowedUSD, and various cumulative measures. In our analysis, a p-value below 0.05 indicates that a series is stationary, while values above this threshold suggest non-stationarity due to trends or shifting variance.
Our ADF tests revealed that many variables, like borrowedUSD, repaidUSD, and interestEarnedUSD, were already stationary. However, key metrics such as totalCollateralUSD, totalUtilizationRatio, and the OHLC price series proved non-stationary (p-value ≥ 0.05). To address this, we applied first differencing (Δyt = yt − yt-1) to these variables, effectively removing the underlying trends. A subsequent ADF test confirmed that the transformed series became stationary (p-value < 0.05), providing a robust foundation for further analysis.
With our dataset properly preprocessed, we conducted a correlation analysis between lending metrics and Ethereum’s volatility. From a comprehensive set of 26 metrics, we selected candidate variables based on their potential impact. The results are summarized below:
Overall, these detailed findings reinforce that repayment activity—especially the numberOfRepays—is a strong and consistent indicator of Ethereum’s volatility across stablecoins. The robust correlations in each ecosystem suggest that monitoring these metrics can provide actionable insights into market stress and liquidity adjustments
When building a larger predictive model, it is critical to account for multicollinearity—the presence of high inter-correlations among independent variables—which can distort coefficient estimates and inflate the variance of predictions. In our analysis of lending metrics, we observe significant multicollinearity across key variables within each stablecoin ecosystem.
The very high correlation (0.837) between numberOfRepays and numberOfWithdraws indicates that these two metrics likely capture overlapping information. This suggests that when traders are repaying loans, they are also frequently withdrawing funds, hinting at similar underlying behavioral drivers. Such redundancy can complicate the interpretation of a model and may lead to instability in coefficient estimates if both variables are included.
In the USDT ecosystem, all three variables exhibit moderately high correlations. The correlation of 0.663 between repaidUSD and borrowedUSD suggests that larger repayment volumes tend to occur alongside higher borrowing activity, indicating intertwined liquidity management strategies. This overlap means that including both might not add substantial new information, and model performance could benefit from selecting only one representative variable or employing dimension reduction techniques.
For DAI, the correlation between numberOfRepays and numberOfWithdraws (0.687) is relatively high, indicating that these metrics move together consistently. Although the correlations with numberOfBorrows are slightly lower, they still suggest moderate multicollinearity. This implies that frequent transactions in repayment and withdrawal are closely linked, potentially reflecting similar user behavior in a decentralized lending environment. As a result, using all three variables in a predictive model might introduce redundancy, and careful variable selection is recommended.
The high levels of multicollinearity across these lending metrics suggest that the variables are capturing similar aspects of market behavior, such as liquidity adjustments and risk management strategies. In constructing larger models, it may be beneficial to use techniques like principal component analysis or regularization methods to mitigate multicollinearity, or selectively choose the most representative metrics. This approach helps ensure more stable coefficient estimates and clearer interpretations of how lending behaviors drive Ethereum’s volatility across different stablecoin ecosystems.
Our analysis in Part 2 builds on our earlier findings by examining the connection between stablecoin lending activity and Ethereum’s volatility. Using the Garman-Klass estimator, we captured a more complete picture of intraday price fluctuations, while applying the Augmented Dickey-Fuller test and first differencing ensured that our lending metrics were stationary and reliable for analysis.
In the USDC ecosystem, we observed a moderate correlation (0.437) between the numberOfRepays and ETH volatility. This suggests that increased repayment frequency—likely reflecting institutional rebalancing and risk management—can serve as an early signal of market stress. Additionally, moderate correlations for both withdrawnUSD (0.361) and numberOfWithdraws (0.357) indicate that significant fund outflows reduce on-chain liquidity, making ETH prices more sensitive to fluctuations.
USDT displays a slightly different profile. Despite lower overall activity, its numberOfRepays exhibits a stronger correlation of 0.491 with ETH volatility, implying that even modest shifts in repayment behavior can have a pronounced impact on market sentiment. Furthermore, moderate correlations for repaidUSD (0.344) and borrowedUSD (0.262) suggest that liquidity adjustments in USDT are driven by both the magnitude and frequency of transactions.
For DAI, although operating on a smaller scale, the numberOfRepays shows a robust correlation of 0.492 with volatility. The ecosystem’s count-based metrics (numberOfWithdraws at 0.344 and numberOfBorrows at 0.255) underscore that frequent, small-scale adjustments effectively signal shifts in market sentiment, even if large capital movements are less common.
Our multicollinearity analysis highlights significant overlaps among key variables, emphasizing the need for careful variable selection in predictive models. Collectively, these findings reinforce that repayment activity is a critical indicator of market stress across stablecoins.
For timely market data and deeper insights into the rapidly evolving DeFi landscape, reach out to Amberdata today.