Analyzing NFT Market Predictability: Factors and Effects
Foreword
As a master’s student in finance and fintech, I have been reading numerous papers on the subject. Recently, I came across an interesting paper that discusses pricing in the NFT market using traditional financial methodology. In this document, I will share some parts of it.
The authors created indices for the NFT market, similar to the real estate market indices. They tested whether the NFT market return was influenced by other markets or assets. Additionally, they examined whether volatility and the NFT valuation ratio significantly predicted NFT market returns in time series, and tested for the existence of the size and reversal effect.
The paper is titled “The Economics of Non-Fungible Tokens”, and the authors are Nicola Borri, Yukun Liu, and Aleh Tsyvinski. It was published in March 2022 and contains abundant and detailed content. The NFT data is rich nowadays, and while I have only shared some topics, it is worth reading the entire paper if you have the time.
It’s the link to this paper：https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4052045
Outline
- Data
- The NFT market index
- Exposure to the NFT Market
- NFT Market Return Predictability
- Conclusion & Useful Tool
Data
The data comes from major exchanges such as Cryptokitties, Gods Unchained, Decentraland, Opensea, and Atomic. The dataset includes only transactions that represent the transfer of ownership of NFTs, excluding transactions for the minting of NFTs and auction bids. The data is divided into 52 weeks. If an NFT is traded more than once a week, then the average price is taken. The duration of the data is from the beginning of 2018 to the end of 2021.
NFT market index
The NFT market index is constructed by using the “repeat sales method”. This method is suitable for items that are heterogeneous and traded infrequently, such as the real estate market.
The authors assume that the difference in logged prices for two NFT sales is equal to the difference in the corresponding logged NFT market index, plus a random error that captures the idiosyncratic component of the specific NFT.
rit’t：return
pit’ ：logged sale price
pit：logged purchase price
bt：logged NFT market index at time t,
bt’：logged NFT market index at time t’
uitt’：multiplicative error term
where t’ > t. And removing return at the 99% level each week, price < 1, transaction < 2.
This figure shows the NFT Market Index, which is constructed by using the baseline repeat sales method. The indices at the beginning are normalized to one.
After understanding the NFT market index, we can look at the summary statistics of the NFT market return：
Panel A presents the overall summary statistics of weekly NFT market returns.
NFTH index：the heteroscedasticity-adjusted NFT index
CMKTRF：the cryptocurrency market excess return
MKTRF：the stock market excess return
Panel B shows the weekly average NFT index and NFTH index return by quarter.
It is evident that the NFT market has a larger volatility, with the standard deviation and mean significantly greater than those of other markets. Moreover, the Skewness is 1.098, indicating a right-skewed distribution and most transactions resulting in negative returns. The Kurtosis is 7.732, indicating a fat-tailed distribution with extremely high returns occurring occasionally.
In panel B, we observe a rapid increase in the index return during 2020Q2, which peaked in 2021Q3. This is consistent with the real market condition based on the real transaction volume.
Exposure to the NFT Market
This section examines the relationship between NFT market returns and the cryptocurrency market, as well as traditional markets such as stocks, commodities, and currencies.
1. Exposures of NFT Market to Cryptocurrency Factors
The authors use three factors to examine the relationship between NFT and cryptocurrency:
CMKTRF：represents the cryptocurrency market,
CSIZE：represents the size, and
CMOM：represents momentum. Additionally,
CVALUE：represents value, which is measured by the price-to-new address ratio.
In all specifications, CMKTRF is significantly positive at the 1% level, even when controlling for size, momentum, and value factors. This indicates that the excess returns of the NFT market are influenced by excess returns of the cryptocurrency market, and in the same direction.
For instance, in column (1), a 1% increase in the excess returns of the cryptocurrency market would lead to a 0.789% increase in the excess returns of the NFT market.
2. Exposures to Traditional Asset Market Factors
In this section, the authors examine the relationship between the NFT market and traditional asset markets, including stocks, commodities, and currencies. To analyze the stock market, they use several models, including the CAPM, Fama-French 3-factor model, Carhart 4-factor model, and Fama-French 5-factor model.
In panel A, the following variables are introduced:
MKTRF：represents the excess return of the stock market
SMB：represents the size effect
HML：represents value
MOM：represents momentum
RMW：represents an investment
CMA：represents profitability
According to this table, the MKTRF factor has a significant positive impact at the 5% level, while the other factors have no effect except for the value factor (HML). This suggests that the NFT market’s excess return is positively influenced by the stock market’s excess return, but the effect is small due to the low R-squared.
The variables in panel B：
Gold
BBG Commodity：the Bloomberg commodity index
Dollar：US Currency
Carry：Carry Trade
In this section, we found that only the commodity index could influence the excess return of the NFT market. Other factors had no significant effect. The coefficient is significantly positive at a 1% level, but the R-square is very small. This factor is effective but not huge.
3. NFT-Related Cryptocurrencies and NFT Market
Some companies issue both cryptocurrency and NFT, such as Decentraland’s MANA token, Axie Infinity’s AXS token, and others.
In this section, we want to examine whether the excess return of NFT-related cryptocurrencies is influenced by their NFT return and study whether the cumulative excess return of NFT-related cryptocurrencies can be predicted by lagged NFT returns.
CMKTRF is the cryptocurrency market excess return.
In Table Panel A, we can see that the NFT market excess return is significantly positive at the 1% level. Even after controlling for the CMKTRF factor, it remains significant at the 5% level. This suggests that the excess return of NFT-related cryptocurrencies can be influenced by NFTs themselves. This result is not surprising and simply confirms my expectations.
Panel B is a test of predictability that uses the lagged NFT market excess return to predict the future cumulative excess return of NFT-related cryptocurrencies. The results show that the NFT market excess return can predict the future two-week cumulative excess return of NFT-related cryptocurrencies. This predictability is significant at the 5% level. Specifically, a 1% increase in NFT market index excess return leads to a 0.33% increase in NFT-related cryptocurrencies' cumulative excess return at the two-week horizon.
NFT Market Return Predictability
This section focuses on predictability. First, we studied some variables that are used to predict time-series returns for traditional asset markets such as volatility, valuation ratio, attention, past returns, and volume.
Second, we studied cross-sectional return predictability and the variables that include size and momentum to test the size effect and reversal effect for NFT returns.
1. Time-Series Return Predictability
a. Volatility
Some papers have shown that volatility tends to negatively predict future asset returns. In this section, we test whether the NFT market exhibits the same effect.
The authors assume that the volatility of NFT market excess returns is the sum of the trailing squared NFT market excess returns of the past eight weeks.
First, Vol represents volatility, We can observe that volatility is significantly negative from the five-week to eight-week horizon. This means that the cumulative excess return of the NFT market can be influenced by volatility in the 8-week horizon. Additionally, if the volatility increases, the future NFT market cumulative excess return will decrease. This result is consistent with previous research.
b. Valuation Ratio
In the equity market, the valuation ratio is typically represented by market-to-fundamental ratios, which are calculated by dividing market value by book value or fundamental value. In the NFT market, the market value of NFTs is determined by the NFT market index (repeat sales index), while the fundamental value is based on the total transaction count.
As a result, the valuation ratio in the NFT market is measured using the logged index-to-transaction ratio.
In Panel A, we can observe that the valuation ratio has a consistently negative relationship with NFT market excess returns across all horizons. This indicates that the valuation ratio is a strong predictor of the market’s performance, as an increase in the valuation ratio is significantly associated with a decrease in cumulative excess returns.
Moreover, the R-squared value is quite high, with a value of 0.269 over an eight-week horizon. This suggests that the valuation ratio is an important factor that can be utilized to predict NFT market excess returns.
c. Attention
In this subsection, we examine whether investor attention can predict future NFT market returns. To accomplish this, the authors create a metric called the “Google_NFT” deviation, which measures the difference in Google searches for NFT in a given week compared to the preceding eight-week average. This metric is calculated by subtracting the average of the preceding eight weeks’ search volume from the search volume in the current week.
In addition to Google_NFT, other attention metrics such as Google_crypto and Google_bitcoin are also considered.
This table confirms that investor attention does not significantly influence future NFT market cumulative excess returns. Therefore, attention is not an important factor in predicting NFT market performance.
In my opinion, I think only using Google search count to proxy attention may not be quietly appropriate, maybe can include the mention count of social media（e.g Twitter, Reddit） or articles to proxy attention！
d. Momentum
The momentum effect has been observed in various markets, such as stocks, bonds, cryptocurrencies, and real estate. In this subsection, we investigate whether this effect is existing in the NFT market.
To accomplish this, the authors use current NFT market excess returns to predict future cumulative returns in the NFT market
Based on the results presented in the table, it is surprising for me to note that there is no momentum effect observed in the NFT market. Therefore, attempting to chase prices in this market is unlikely to result in profits.
e. Volume
In most markets, there is a positive relationship between trading volume and returns. In this subsection, we aim to examine this relationship in the NFT market.
To achieve this, we construct a volume metric by measuring the deviation of NFT trading volume in a given week from the average of the preceding eight weeks.
The table reveals a significant and positive relationship between trading volume and future cumulative NFT market excess returns over one to three-week horizons.
This result aligns with my expectations, as similar findings have been observed in other markets. Moreover, it is consistent with the demand-supply theory, which suggests that the price is influenced by the traded quantity.
However, in panel B, when the valuation ratio and volatility are added as additional factors, the effect of volume becomes considerably weaker. Nevertheless, the valuation ratio remains a significant predictor of NFT market returns.
In conclusion, the factors of volume, volatility, and valuation ratio (logged index-transactions) are all important in predicting NFT market returns. However, the most crucial factor is the valuation ratio (logged index-transactions), and it can be the primary focus when predicting NFT market returns in time-series analysis.
2. Cross-Sectional Return Predictability
In this section, we investigate the presence of the size and reversal effects in the cross-section of NFT returns. This analysis involves examining the differences between individual returns.
1. Size Effect
Previous research has shown that in the cross-section of stock returns, smaller stocks tend to outperform larger stocks. This phenomenon is commonly referred to as the size effect.
The authors use the market price of individual NFT（(t’- t) lnPit）to identify the potential size effect （higher means big size）and control the market index（bt）to form the equation. γ is the elasticity, which means when the purchase price changes 1% then the return will change γ percent. So we test whether γ is significant, if it’s then the size effect is present.
In this table, we can find the coefficient is -0.004 in Full Sample, and the result is significantly negative. That means when the purchase price increases 1% then the average NFT return will decrease 0.004%, and the result suggests that expensive NFTs tend to significantly underperform when compared to the less expensive NFTs. （The possible reason is explained in the paper on page 25.）
2. Reversal effect
The momentum and reversal effects are commonly examined in asset pricing. In this section, we test these effects using past returns. Additionally, we only consider NFTs that have had at least three transactions.
ri,b is the logged average weekly return of NFT i’s previous repeat sales, and γ is the elasticity.
The results show a significant negative coefficient for past weekly returns in the full sample. Specifically, a 1% increase in past weekly returns results in a 0.014% decrease in future returns. This confirms the existence of the reversal effect, where high average past returns tend to underperform significantly.
Conclusion
In this study, the authors created an NFT market index using the repeat sales method and found that the cryptocurrency market had the most significant impact on the NFT market.
They also examined the predictability of the NFT market return and found that the most important factor in time-series return predictability is the valuation ratio.
In addition, they confirmed the existence of the size effect and reversal effect in the cross-section return predictability. Specifically, small-size NFTs (low price) outperformed big-size NFTs (high price), and high past returns significantly underperformed in the future.
The paper provides in-depth explanations and details about the methodology and factors used. Therefore, it’s worth reading the whole paper to better understand the research.
- All the figures and tables are from the original paper.
BTW, if you are interested in NFTs, I ‘d like to recommend a useful tool to you.
https://dune.com/corncobbb/swf-lab-nft-analysis
It’s the NFT analysis Dune Dashboard, which provides an overview of the NFT market as well as analyses of price, volume, profit, holders, and wash trades. All you need to do is enter the NFT contract address in the space provided in the upper left corner.
I think it’s helpful for your NFT analysis！！
