None of the seven activity measures Brigida tests makes TVL-to-market-cap a signal worth defending at a risk meeting. DEX volume produces the lowest full-sample value-weighted p-value, 0.048. Adjust for multiple comparisons across the six on-chain tests and it rises to 0.291. Brigida acknowledges the failure: his abstract reports several rejections at unadjusted levels, with none surviving Holm adjustment.
The question is how much that null tells us. In the DEX sort, the smallest of the six cells contains a median of one token. The median largest single weight across cells reaches 100%. Concentration aside, the two traded legs capture main effects; neither isolates the TVL interaction discussed in the conclusion. The abstract ends by saying the evidence does not show ecosystem activity reliably turning TVL into an incremental return signal. These legs could not measure that transformation directly. The conclusion therefore reaches less far than it first appears.
From deposited capital to a trading signal
Brigida starts from his own 2025 finding that the crypto market portfolio spans portfolios sorted on total value locked alone. TVL measures deposited capital and follows ecosystem scale, making it a weak proxy for tokenholder cash flow. Activity might distinguish productive capital from capital sitting idle. Consider two chains with identical TVL relative to market cap. Users, developers, fees and DEX volume support the deposits on one chain; the other lacks them. The proposed return advantage belongs to the first token if the market prices that difference slowly.
The weekly test covers January 3, 2022 to July 28, 2025. Tokens qualify for the candidate universe by appearing among roughly the largest 100. CoinMarketCap supplies prices and market caps. DefiLlama supplies TVL, DEX volume and fees, while Artemis supplies active addresses, new wallets, core developers, transactions and new X followers.
At each weekly sort, the median of V = TVL/MCAP divides tokens into lower and higher TVL-ratio groups. All inputs use end-of-week t-1 data. An independent split at the 30th and 70th percentiles of an activity measure, itself divided by market cap, creates six cells labelled SL through BH. S and B denote smaller and larger V. Lagged market cap determines the weights within each cell. Price changes determine returns, excluding issuance and burns from performance.
The cells produce two traded legs:
- LH_TVL subtracts the three high-V cells from the three low-V cells. Its long side holds the low TVL ratio.
- HML_F subtracts the average of SL and BL from the average of SH and BH. It measures high-minus-low activity averaged across the TVL groups.
Each leg is regressed on a market factor, the cap-weighted universe return less the weekly T-bill, and SIZE, the smallest market-cap quartile minus the largest. The joint null sets both alphas to zero. A Bartlett HAC Wald test uses bandwidth 4 and a χ²(2) reference distribution. Holm adjustment covers 6 full-sample tests or 7 common-sample tests.
Do these legs measure an interaction?
Neither leg tests whether activity strengthens the TVL effect. Brigida calls HML_F a high-minus-low activity leg conditional on the TVL ratio. The description has a limited justification: the activity spread is calculated within each TVL group before those spreads are combined. To measure the interaction directly, a leg would take their difference, for example (SH minus SL) minus (BH minus BL). HML_F averages them and cannot identify that difference.
The joint test asks whether either a low-TVL tilt or an activity tilt earns alpha. Yet the conclusion discusses TVL interactions, which neither leg isolates. Answering that question requires a third leg absent from the paper.
The DEX results make the distinction concrete. Full-sample LH_TVL alpha is 0.798% per week (SE 0.393); HML_F alpha is 0.883% (SE 0.475). Those estimates leave open whether activity changes the TVL premium or contributes a separate tilt.
A DEX cell can be a single token
Brigida reports the concentration himself. Across the activity measures, median eligible universes range from 16 tokens for new wallets to 54 for developers. The median smallest cell contains 1 token for DEX and 2 for New X. Taking the largest weight in any cell each week, its time-series median ranges from 79.2% for developers to 100% for DEX. A median 100% allocation to one DEX name leaves the DEX alpha largely dependent on that token's return.
Equal weighting offers an obvious remedy, but it changes which sorts reject. Fees reject under value weighting in the common sample at p = 0.012; equal weighting gives 0.248. For common-sample DEX, equal weighting produces p = 0.007, rising to 0.051 after Holm, with a bootstrap value of 0.050. Equal-weighted transactions reject over the full sample at 0.014, adjusted to 0.086. The paper acknowledges both the changing identity of the lowest-p sort and the absence of family-wise evidence at 5% under any specification.
The DEX and developer sorts exclude Bitcoin. Brigida explains that these measures have a different economic interpretation for Bitcoin than for smart-contract ecosystems. Including it in DEX lowers the full-sample p from 0.048 to 0.015. Across 78 of the 83 common weeks, the p is also 0.015, alongside a bootstrap value of 0.050. The paper attributes the change to Bitcoin's dominant weight and atypical DEX activity. It accordingly urges caution about isolated DEX rejections.
Large X-follower alpha, little power
New X followers are available from August 2023, leaving 97 usable weeks. The common sample contains 83 weeks beginning September 4, 2023. The activity leg earns an estimated alpha of 1.766% per week, with a HAC SE of 0.751. Its joint-test p is 0.052 asymptotically and 0.106 with the four-week block bootstrap; Holm takes it to 0.262. Only fees has a common-sample bootstrap p below 5%, at 0.038, and its Holm value is 0.086. Testing fees for alpha stability yields 0.031 before adjustment and 0.188 afterward.
A 1.77% weekly estimate on 83 observations, with that standard error, chiefly tells us how little the test can distinguish. And 499 bootstrap draws leave a p of 0.050 too noisy to place confidently on either side of the threshold. We raised the same concern about short windows in weekend crypto variance.
Replication decisions left open
The universe includes tokens appearing among approximately the top 100 during the period. That wording permits membership to depend on subsequent rankings. A replication should instead reconstruct eligibility each week using that week's top 100. We did not find a statement establishing whether the DefiLlama and Artemis histories were preserved as first published or later revised. Backfilled TVL makes that distinction consequential. We also did not find the token count within each SIZE quartile. Nor did we find costs, turnover or slippage. Every alpha quoted here is therefore gross and long-short, with weekly rebalancing in cells that can contain one token.
We could not run this ourselves.
Our data lacks historical TVL, the DefiLlama and Artemis activity series, and point-in-time market caps for a top-100 universe. Our crypto price coverage extends to about 50 tokens.
Brigida states his conclusion carefully. The result "does not imply that TVL or ecosystem activity lacks economic information," he writes; the information simply does not appear as abnormal returns against market and size. Every alpha is gross, and no specification survives Holm. That is enough reason to keep capital away for now.
The author calls for longer attention histories, broader token coverage and richer benchmark models. His requests leave out an explicit interaction leg and net-of-cost returns. One result would change our mind: a positive interaction leg, net of costs, on data after July 2025, using a universe reconstructed weekly from that week's top 100 and extending well beyond 16 to 54 names.