Beyond Beta: Decomposing the Fundamental Drivers of DeFi Token Returns Across Market Regimes

1. Introduction

In traditional financial markets, asset prices are heavily tied to company fundamentals. Earnings reports, credit rating changes, and dividend announcements are all events in which the public gains new information about the underlying health of a company, and markets tend to react swiftly and decisively.

In decentralized finance, this data is neither private nor temporally sparse; it is published on-chain, at high frequencies, and available to anyone with an internet connection. In theory, DeFi token prices should be tied to protocol fundamentals like revenue generation and protocol usage, however in practice, they are widely observed to be heavily correlated with the prices of benchmark tokens like Bitcoin, Ethereum, and Solana (Șoiman et al., 2022). This raises the question of whether protocol fundamentals are meaningfully priced into token returns, or if returns are largely at the mercy of market conditions.

Most research on this topic largely focuses on the correlation between returns of benchmark tokens and DeFi tokens but stops at that. This paper moves beyond correlation towards attribution by quantifying the relative contribution of each protocol fundamental and of market beta through Shapley value regression. By examining tokens across four DeFi sectors — decentralized exchanges (DEX), lending, perpetuals, and a miscellaneous category — this study tests how attribution patterns differ across sectors. Additionally, it measures the difference in attribution between bullish and bearish market regimes, providing insight into how attribution patterns differ across regimes.

2. Theoretical Framework

DeFi token returns are theoretically determined by two distinct components: exposure to systematic market risk (market beta), and the fundamental performance of the underlying protocol. The market beta component has been thoroughly investigated by past studies. Șoiman et al. (2022) find that cryptocurrency market conditions dominate all other drivers of DeFi returns, establishing the overall health of the crypto market as a meaningful predictor. This study examines the less researched fundamental component and decomposes it into four main categories: those that capture protocol demand, engagement, output, and scale.

The specific fundamentals that fit these categories will differ between sectors. For example, lending loans are a good measure of engagement for lending protocols, but are nonexistent for DEX protocols. Other fundamentals, such as daily active users (DAU), are more generalizable across sectors. The specific fundamental variables used to span each category are discussed in detail in the data section, where sector-specific measurement decisions are also justified. Regardless of sector, the theoretical justification for why fundamentals should matter mirrors that of traditional finance.

For example, a protocol’s fees generated are analogous to a company’s earnings and DAU reflect user adoption and retention, the same signals used in equity markets to forecast future growth. These on-chain metrics are essentially DeFi’s version of traditional finance valuation metrics. However, a key distinction from TradFi is that crypto markets may operate as a reflexive feedback loop, where the causal direction between price and fundamentals is less clear. In equities markets, earnings improve, new users are attracted, revenue increases, valuations become more favorable, and investors buy the stock. In crypto markets however, this loop can sometimes reverse. Sometimes a token will pump from social media hype, drawing attention that leads to new users, leading to greater fee generation. This means that fundamental changes are sometimes downstream of price changes, rather than being an explanatory driver of token price.

In support of this idea, Xu et al. (2025) apply DCF analysis to DeFi protocols, finding that tokens trade at significant premiums to their fundamental value, suggesting that market prices are not fully anchored to on-chain fundamentals, though whether this holds true across sectors and market regimes remains an open question.

To further examine this relationship, this study will investigate three hypotheses:

• H1: DeFi token returns are primarily driven by systematic market conditions.
• H2: Protocol fundamentals contribute incremental explanatory power beyond market conditions.
• H3: The relative contribution of fundamentals vs. market conditions differs across bull and bear market regimes.

3. Data

To quantify systematic market conditions, daily closing prices for Bitcoin (BTC), Ethereum (ETH), and Solana (SOL) were retrieved using the tidyquant package in R. BTC and ETH were selected due to their dominance in crypto markets: they are consistently some of the top ranking coins by trading volume and market cap, and because of this, they serve as a proxy for the health of the overall crypto market. SOL was included given its role as a primary infrastructure layer for many DeFi protocols. The sample period begins December 1st, 2024, reflecting the earliest date for which complete on-chain fundamental data was available across all protocols in the sample. All on-chain fundamental data was retrieved from Artemis, a crypto and equities data aggregator.

Three main DeFi sectors, DEX, lending, and perpetuals, were selected for analysis as they represent the main body of the DeFi ecosystem. Within these sectors, six DeFi tokens were selected for analysis based on their relative market cap and on consistency of data availability. For DEX protocols, Uniswap (UNI), PancakeSwap (CAKE), and Aerodrome Finance (AERO) were selected, and for lending protocols, Aave (AAVE) and Morpho (MORPHO) were selected. The perpetuals sector is represented solely by Hyperliquid (HYPE), reflecting both its dominant position within the sector and the absence of comparable data availability among competing protocols.

For each sector, the daily prices were extracted, as well as four fundamentals to represent protocol demand, engagement, output, and scale. For DEX, spot volume, DAU, capital efficiency, and total value locked (TVL) were selected to represent these categories respectively. For lending protocols, lending loans, lending deposits, fees, and TVL were selected. For perpetuals, volume, DAU, revenue, and TVL were selected.

Additionally, two protocols were selected to capture mechanisms outside the three primary sectors: ether.fi (ETHFI), a liquid restaking protocol with a crypto card product, and pump.fun (PUMP), a token launchpad. These protocols were included to test whether patterns differ for structurally distinct protocol models. For ETHFI, card volume, number of cards, fees, and TVL were selected, and for PUMP, tokens graduation, native buybacks, revenue, and launchpad volume were selected. PUMP is the only token in this sample that starts on July 22nd, 2025 rather than December 1st, 2024 because this is the earliest point at which token price data is available.

4. Methodology

To decompose the relative contribution of protocol fundamentals and systematic market conditions to DeFi token returns, this study estimates token-level OLS regressions for each of the eight protocols in the sample, separately for bull and bear market regimes.

First, the proxy for underlying crypto market health was constructed using a Principal Component Analysis (PCA) of BTC, ETH, and SOL returns. These three tokens are highly correlated, meaning that including them individually would introduce multicollinearity into the model, biasing OLS estimates. PCA extracts a single composite factor — the first principal component — that captures the shared variation across all three, serving as a cleaner proxy for systematic market conditions.

Bull vs. bear market classification was performed using the BTC 200-day moving average, signifying a bull market if BTC is above its 200 day MA, and a bear market if it is below. This is a widely adopted threshold for distinguishing sustained trend regimes in traditional financial markets, and is commonly applied to cryptocurrency as well.

For each token the closing prices were used to calculate log returns by Equation 1, where is the closing price on day t, and the daily log change in each fundamental was calculated by Equation 2, where is the fundamental value on day t.

Log differences were used over simple Δ values to induce stationarity in the price and fundamental data. All data were then aligned by date, and missing or zero values were removed prior to log transformation.

This gives a total of 300 daily observations for PUMP (99 Bull, 201 Bear), and 533 daily observations for the remaining tokens (297 Bull, 236 Bear). Next, robustness testing in the form of an Augmented Dickey-Fuller (ADF) test was carried out to ensure that all token time series data remained stationary. All tokens return an ADF test p-value less than 0.05, meaning that none of the time series display non-stationarity, as desired.

Next, ordinary linear regressions were run on each token’s two separate time series’, split by bull and bear market classification. For each token i, and fundamental F, the following regression equations were calculated:

To account for heteroskedasticity, HC3 robust standard errors were employed across all models. Breusch-Godfrey tests revealed serial correlation in the CAKE and MORPHO models, for which Newey-West HAC standard errors were substituted, correcting for both heteroskedasticity and autocorrelation.

Finally, in order to determine the relative contribution of each variable from the regressions, Shapley Value Regression was implemented. One of the weaknesses of OLS is that it assumes the predictor variables are uncorrelated, however with fundamental data, this is usually not the case. For example, if DAU is increasing, it is reasonable to assume that fees are increasing as well. In the presence of multicollinearity like this, OLS coefficient estimates break down because the model cannot distinguish between each variables individual contribution. Shapley regression resolves this by evaluating each predictor’s marginal contribution across all possible variable orderings and averaging the result. Since each variable’s contribution is averaged across all possible orderings, no single variable can claim the full credit for variance that it shares with a correlated predictor, producing an attribution of R² that is robust to correlation among predictors. Shapley Regressions were applied to each token’s bull and bear regression’s R², and heat maps were constructed to display each fundamental’s relative contribution to price movement.

5. Results

Across tokens and market regimes, OLS regression reveals the PCA constructed market beta factor to be the most consistent and statistically significant predictor across the board. OLS results are displayed in tables 1-5, note that HC3 robust standard error estimates are displayed in parentheses. Fundamental variables are largely insignificant at the individual coefficient level, though R² values ranging from 0.34 to 0.67 suggest the model captures meaningful variation in returns. Shapley decomposition is used to quantify the relative contribution of each variable to this explained variance.

Heat maps are constructed by sector to display relative contributions from Shapley Value Regression. ETHFI and PUMP are displayed on separate heat maps since different fundamentals were used for their analyses.

On average, DEX tokens returns seem to be the most dependent on market beta, with all three tokens displaying over 50% of variance defined by market beta, regardless of regime. In contrast, lending protocols seem to go against the notion that token returns are largely tied to the overall market health, with MORPHO being the clear standout. There does not appear to be a consistent pattern of tokens being more market dependent during bear regimes, with about half of the tokens actually showing more fundamental dependence during bear regimes.

6. Discussion

The DEX sector gives our most clear confirmation of Hypothesis 1, with UNI being the clearest example. In bull markets, nearly 92% of their price variance is determined by market beta, and this only drops to around 86% in bear markets. CAKE and AERO show less of a lockstep pattern, but their variance is still majority determined by market beta, regardless of market regime. Fees appear to be the most informationally relevant fundamental for these two tokens returns. Additionally, DAU and TVL seem to be mostly meaningless. TVL was included as a scale metric to maintain consistency across sectors, but its theoretical relevance for DEXs is weaker than other sectors, so this result is expected. The low Shapley value for DAU is potentially due to the number of automated transactions that occur through DEXs, meaning that DAU is not actually meaningfully measuring human engagement with the platform.

One potential reason for why we see UNI much more closely tied market beta is because of its token structure. UNI’s token is used for protocol governance, making it more of a measure of investor sentiment towards DeFi rather than tied to value accrual. If people are excited about DeFi — and crypto in general — they are more likely to buy a governance token, just like they are more likely to buy BTC, ETH, or SOL. In contrast, AERO operates a vote-escrow model in which token holders lock AERO to direct liquidity emissions and earn a direct share of trading fees, creating a mechanical link between protocol performance and token value. CAKE similarly offers governance rights but more importantly distributes protocol incentives to holders. Both tokens therefore carry a more direct claim on protocol value accrual than UNI, which likely explains their greater fundamental sensitivity.

The lending sector acts in stark contrast to the DEX sector, providing counter-evidence for Hypothesis 1. MORPHO shows virtually no dependence on overall market health regardless of regime, and AAVE shows a similar pattern but displays more market dependence with 34.2% of variance determined by market beta in bull regimes. Interestingly, AAVE displays less beta dependence in bear markets, a counterintuitive finding that will be discussed further with HYPE. The attribution of variance between fundamentals for these tokens gives insight into how consumers value lending protocols. We see immediately that fees hold almost no weight, signifying that investors either do not pay attention to or do not care about the fees, and therefore revenue generated by lending protocols. Lending loans most likely contribute less than lending deposits because they actually depend on lending deposits. Without deposits and liquidity, no loans can be given out. The most weight is given to lending deposits and TVL, which both signal future capacity and protocol health. This indicates that investors value the stability and projected future growth of lending protocols more than they value what the protocol is doing now.

Similarly to AAVE, HYPE shows significantly more fundamental dependence in bear markets than in bull markets. In a bear market, we expect fear to take over and altcoin selloffs to follow major benchmark coin selloffs with less consideration for individual protocol health. AAVE and HYPE seem to do the opposite: when in a bull market, they show drops in market health dependence of about 20% and ride the wave, with investors focusing less on fundamentals. HYPE’s perp DAU shows the biggest fundamental variability between regimes, with attribution increasing from 3.9% to 9.1% when going from bull to bear. This signifies that in bear markets, investors are paying closer attention to user retention and rewarding consistent activity.

ETHFI shows an interesting split of attribution between market beta and TVL in bull markets, however this disappears in bear markets with much of the TVL attribution disappearing, and market beta accounting for most of returns variance. For ETHFI, TVL represents Ethereum staked through the protocol, meaning that when markets sell off, this staked ETH is illiquid, making TVL an inflated measurement of scale during selloffs. Therefore, the markets stop treating it as a measure of current protocol health. Number of cards and card volume both provide negligible variance in returns, signifying that markets are not yet pricing these in, or that the market doesn’t believe that the debit card product performance is indicative of long term value accrual.

PUMP displays low dependence on market beta, with only 19% of return variance attributed to market conditions in bull regimes and 21.4% in bear regimes. Instead, revenue and native buybacks together account for roughly 80% of explained variance across both regimes, making PUMP one of the most fundamentals-driven tokens in the sample. Notably, launchpad volume itself contributes almost nothing, suggesting that investors are not rewarding raw launch activity but rather the revenue PUMP extracts from it. The stability of this attribution pattern across regimes indicates that PUMP’s fundamental signal is not regime-dependent, but that investors appear to price revenue and buybacks consistently regardless of broader market conditions.

7. Limitations

While the findings presented above offer meaningful insight into regime-dependent return attribution across DeFi protocol categories, several limitations of this study warrant consideration.

The small number of tokens within each sector limits the generalizability of sector-wide findings. With three DEX tokens, two lending tokens, and a single perpetuals representative, observed patterns may reflect characteristics of the selected protocols rather than sector wide patterns. For example, HYPE’s bear-regime fundamental sensitivity may be specific to Hyperliquid’s dominant market position rather than indicative of perpetuals protocols broadly. While the sample size was chosen due to data availability, future research should attempt to expand the token sample within each sector.

This model only captures on-chain fundamentals, meaning that it misses any price action from off-chain events such as new protocol announcements, token drops, or partnership news. This means that any price movement by these events is misattributed to other factors, likely inflating the market beta attribution since this is meant to capture any movement not explained by the fundamentals.

Selected metrics may also not be perfect representations of demand, engagement, output, and scale. Future research should test fundamental attribution hypotheses using different sets of fundamental data for sensitivity analysis.

Regime classification relies on the BTC 200-day moving average, a widely adopted threshold for distinguishing sustained trend regimes in both traditional and crypto markets. Alternative thresholds such as the 50 or 150-day moving average could produce different regime splits, and future research should perform analysis using alternative thresholds to see how findings are affected.

Lastly, this study uses a relatively short sample period in an attempt to align all tokens from a universal starting date, making regime changes consistent across tokens. While this is useful for this specific model, it only covers roughly 18 months of trading days. Longer sample periods will provide a more robust regime classification as well, with more sample days given for bear regimes in particular.

8. Conclusion

This study decomposed the drivers of DeFi token returns into market beta and protocol fundamentals across bull and bear regimes using Shapley value regression on eight tokens spanning DEX, lending, perpetuals, and structurally distinct protocol categories. The results reject hypothesis 1 as a universal claim. Market beta dominates returns for DEX tokens, PUMP, and ETHFI in bear regimes, but contributes negligibly for lending tokens, where fundamentals account for the majority of explained variance. Additionally, findings offer a more nuanced picture for the second and third hypotheses.

Market beta is the single largest contributor to return variance for six of the eight tokens examined, consistent with prior findings that systematic crypto-market conditions dominate DeFi return dynamics. However, the magnitude of this dominance varies substantially across tokens, suggesting that the conventional framing of DeFi tokens as uniformly beta-driven may be over generalized. In particular, the lending tokens examined show fundamental attribution shares that exceed market beta, with lending deposits and TVL carrying most of the explanatory weight.

The bull vs. bear regime comparison reveals no consistent directional pattern. The intuitive assumption that bear markets mask fundamental signal and amplify selloffs holds for some tokens, but reverses for others. Whether this reflects genuine investor behavior, characteristics specific to these protocols, or sampling noise cannot be resolved with the current sample and warrants further investigation.

These findings should be read as token-level observations rather than sector-level conclusions. With at most three tokens per sector, the patterns observed of DEX tokens as beta-driven and lending tokens as fundamentals-driven are suggestive hypotheses for future testing on broader samples, not established sector characteristics. Similarly, the explanations for why specific tokens behave as they do (governance vs. value-accrual token structures, illiquid TVL during selloffs) are constructed with model results in mind and would need to be tested directly, rather than inferred from a handful of cases.

The broader implication is that the question “are DeFi tokens priced on fundamentals?” has a nuanced answer. Attribution depends on the token’s value-accrual mechanism, the sector’s economic logic, and the market regime. Any framework that treats DeFi tokens as a homogeneous asset class will miss most of what’s going on. This implies that the explanatory power of fundamentals varies across tokens. For researchers, it argues for designs that model individual token level behavior, rather than pooling across the asset class.

1. Șoiman, Florentina, Jean-Guillaume Dumas, and Sonia Jimenez-Garces. (2022). “The Return of (I)DeFiX.” Université Grenoble Alpes. arXiv:2204.00251. https://arxiv.org/abs/2204.00251
2. Machart, Felix, and Mateusz Dominiak. (2025). “Do Fundamentals Drive DeFi Valuations? Analyzing Evolving Market Efficiency with Random-Forest Models (2021–2025).” Greenfield Capital Research. https://greenfieldcapital.com/2025/11/21/defi-fundamentals-report/
3. Xu, Teng Andrea, Jiahua Xu, and Kristof Lommers. (2025). “DeFi versus TradFi: Valuation Using Multiples and Discounted Cash Flows.” arXiv:2210.16846v2. https://arxiv.org/abs/2210.16846
4. Dancho, Matt, and Davis Vaughan. (2026). tidyquant: Tidy Quantitative Financial Analysis. R package version 1.0.12. https://CRAN.R-project.org/package=tidyquant
5. Mayer, Michael, and David Watson. (2025). kernelshap: Kernel SHAP. R package version 0.9.1. https://CRAN.R-project.org/package=kernelshap
6. Wickham, Hadley. (2016). ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag.
7. Mayer, Michael. (2025). shapviz: SHAP Visualizations. R package version 0.10.3. https://CRAN.R-project.org/package=shapviz
8. Artemis Analytics. (2026). Artemis Sheets [Data retrieval plugin for Google Sheets]. Accessed July 30, 2026. https://www.artemis.xyz

I, Xavier Vortigern, certify that this submission is my own original work prepared for the Ledger N3XT Research Competition, that it has not been previously published, and that all data sources, methods, and prior research referenced herein have been properly cited.

Xavier Vortigern  ·  September 2026