Pocket

##### ABSTRACT

The current popularity of non-fungible token (NFT) markets is one of the most notable public successes of blockchain technology. NFTs are blockchain-traded rights to any digital asset; including images, videos, music, even the parts of virtual worlds. As a first study of NFT pricing, we explore the pricing of parcels of virtual real estate in the largest blockchain virtual world, Decentraland; an NFT simply termed LAND. We show a LAND price series characterised by both inefficiency and a steady rise in value.

##### Findings

Fig. 3 reports the core market efficiency findings. These show a rolling window of p-values for the AVR, AP, and DL tests. We allow a 40-week rolling window to balance the need for sufficient data with the limited time period we have available. We also summarise, in Table 2, the results based on weekly and daily data, and for two daily sub-periods. In Fig. 3, times of market inefficiency are shown by the p-value being below 0.05. While there are more periods where the market is not inefficient in the second half of the chart, the picture, consistent across all measures, is that pricing is generally inefficient. We also note that the AP test shows the smoothest pattern over time, perhaps because of its better suitability for smaller samples. In Fig. 4, we add to the efficiency exploration by charting Hurst scores in rolling 40-week windows. The chart shows a dividing line between above and below 0.5. A value above 0.5 indicates positive autocorrelation – that positive or negative returns tend to be followed by the same direction of return. Values below 0.5 indicate switching behaviour, with high return weeks followed by low return weeks. The findings are interesting because we see for much of the time period, switching behaviour dominated, but since the beginning of January 2021, there has been significant positive autocorrelation. Table 2 confirms the Hurst scores as the most informative measure, with the AVR, AP, and DL tests at the aggregate level simply confirming market inefficiency. For Hurst, at the aggregate level, we see a score reasonably close to 0.5 for weekly data, but there is clear evidence of switching behaviour at the daily level.

##### Conclusions

In this exploratory study, we have introduced pricing behaviour in the rapidly growing market for NFTs. Our initial finding is of inefficiency in pricing, but despite this, a rapid rise in value. This inefficiency is not necessarily surprising. Early-stage markets tend to be driven by a volatile search for suitable pricing models and only slowly emerging market efficiency (Khuntia and Pattanayak, 2018). Our results are probably likely to even flatter on how efficient NFT markets are. For example, there might be market manipulation in pricing, or other fraudulent behaviour. Fraud is an impactful part of cryptocurrencies (Grobys, 2021), so it might also be a part of current or future NFT markets. There are some limitations to the study. The relatively small number of trades we have available is a clear limitation, albeit that this is a necessary feature in early studies. We also have not taken account of the differences between plots of land (location, nearby amenities), which will be important in pricing. This omission is also a factor of the limited number of transactions that are available. The prospects for future studies are potentially limitless, as at the beginning of any new market. An obvious further question is whether there is connection between NFT pricing and cryptocurrency pricing. That is the subject of a companion subsequent study (Dowling, 2021). We could also consider connections to wider asset markets, including stock and bond markets. Another interesting prospective study is on whether there is a fundamental model driving price determination in NFTs. There are also many more NFT markets outside of the single market studied here, so expanding to these new NFT markets would be highly interesting. Something broader to consider is that the asset we track, virtual land, at least has a vivid physical world equivalent in the real estate market, to guide decision making. One wonders what pricing models and physical world learning can possibly apply in the currently most valuable NFT market; the CryptoPunks market of unique and absurd digital comic characters. Fig. 3.Rolling window (40-week) p-value results from automatic variance ratio (AVR), automatic portmanteau (AP), and consistent test (DL) tests of Decentraland LAND NFT market efficiency. M. Dowling

Pocket