The Predictive Power of TVL on L1 Cryptocurrencies
Preface
Previously, I touched on how TVL (Total Value Locked) of a blockchain was related to its MC (Market Capitalization) here. Though there seemed to be a weak relationship between two variables, the effect was quite meager; in other words, I cannot really place my bets on the TVL/MC metric to make money. I were able to do so, it would make Fantom the most undervalued blockchain in existence.
Soiman et al. (2022) wrote their paper on the predictive power of TVL on DeFi tokens — they found out that the overall market sentiment had a much larger effect on DeFi token prices than the TVL of the protocol, and they could not really find the evidence of TVL/MC metric’s predictive power.
I will extend the scope of the research a little bit from DeFi tokens to L1 cryptocurrencies and see if the TVL increase of a L1 chain results in a positive change in the L1 cryptocurrency price.
Method
The research method is pretty straightforward. I scraped the price data of 11 L1 cryptocurrencies from Yahoo Finance, and downloaded the TVL data of 11 L1 chains from DeFiLlama.
I will take the weekly log returns on both time series data, and lag the TVL data by 1 to 4 weeks to see if the change in TVL gives any predictive signs on the future change in price.
To put simply, I will check:
- If L1 TVL rises 10% today
- Does L1 price rises by x% after 1 to 4 weeks?
L1 chains I have investigated includes: ETH, LUNA, BNB, AVAX, FTM, SOL, TRX, MATIC, ONE, RUNE, CELO.
I chose these 11 L1 chains because these are the only chains that offer 1 year+ data on DeFiLlama.
Data
Ethereum
Rows represent the lagged weeks, and columns the test statistics. For Ethereum, the only statistically significant relationship (p < 0.01) comes from lag = 4, which means
“If Ethereum TVL increases by 10% today, ETH price is likely to increase by 10% * 0.3858473 after 4 weeks.”
For 1–3 weeks, the p-values are too high, though p-value of 0.167647548 for lag = 1 might worth take a look.
Terra
For Terra ecosystem, the results are largely the same — lag = 4 is the only significant one at 5% level with p = 0.02428058. Among other 3 lags, lag = 1 has the smallest p-value of 0.79, which practically is meaningless here.
BSC
For BSC, we have lag = 1 as the significant one.
The statistics were quite intriguing, so I drew a simple graph on both time series — and for BNB, the price change were remarkably tied to TVL change.
Compare this to LUNA:
where relationship looks to be nonexistent.
Avalanche
For Avalanche, lag = 1 & 2 are both statistically significant.
Fantom
For Fantom, none are significant. This is coherent with our empirical analysis: We all know Fantom price consistently underperforms its on-chain activity. So for Fantom, even if TVL increases, we do not expect the FTM price to live up to it.
Solana
For Solana, we have lag = 1 & 4 to be significant.
Tron
For Tron, we have lag = 1 to be significant.
Polygon
For Polygon, we have lag = 1 & 2 to be significant at 1% level, and lag = 3 to be slightly over 10% level.
Harmony
For Harmony, there is no relationship whatsoever.
ThorChain
For ThorChain, again there is no relationship. It was a little bit surprising, as ThorChain is known for its TVL * 3 tokenomics. Maybe the fluctuation on the speculative bubble is much greater than the deterministic price of RUNE, so that we cannot predict RUNE price based on the TVL.
Celo
For Celo, there is no significant relationship.
Conclusion
We look at 11 chains to find the predictive power of TVL on MC, but we could not find any recurring relationship between the two variables. For some chains, the relationship existed, but to a varying degree and with a varying lag.
However, it does not mean TVL is a meaningless metric; it simply means that TVL is not useful as a predictive variable. If I regress PriceChange ~ TVLChange directly without introducing any time lags, most of L1 chains are significant at 1% level except for LUNA (p = 0.2), FTM (p = 0.5), and MATIC (p = 0.8).
To sum up, for some chains, TVL growth does not have much impact on its price growth, but for most chains, two are related to some extent. It would be a dangerous decision to make an investment solely based on the TVL metric, but a strong growth in TVL should be a good indicator (though not leading) of the chain growth.
Reference
- https://arxiv.org/abs/2204.00251
- DeFiLlama.com
- Finance.yahoo.com
Join Coinmonks Telegram Channel and Youtube Channel learn about crypto trading and investing