4.1 Iterative Construction of Taxonomies, Indices & Score Modifiers

Based on the asset characteristics discussed in Section 1.3 and the methodologies employed by Bailey, Nickerson and Prat in Section 2 and Section 3, the author has constructed a TokenSpace named TS10 and three hybrid categorical and quantitative taxonomies with weighted scores based on the meta-characteristics, dimensions and characteristics discussed in Section 3.2 paying heed to the design considerations outlined in Section 3.3. These taxonomies are shown in Tables 9, 10 and 11 (contained in this PDF). In applying the Nickerson methodology (Section 3.2.1) some judgement was used to reduce the complexity of the putative taxonomies by consolidating a number of similar and / or overlapping categories into “indexed” ranged score modifiers as discussed in Section 3.3.4. Care was taken to ensure the correct “polarity” of outputted scores - for example a network perceived to have poor stakeholder balance would lead to an increase in Securityness but a decrease in Moneyness. This is consistent with the design goals of being useful, straightforward to apply and minimising arbitrary elements. It is a potential goal for extensions of this work to more thoroughly delineate the impact of each of these elements in finer granularity and assign appropriate score modifiers and / or branch weightings to them.

TS10 Taxonomy Development Iterations:

1) Progression from intuitively reasoned shortlists in Section 3.3 to categorical & indexed dimensions.

2) Assigned unstandardised characteristic score modifiers (weightings incorporated), reduced number of dimensions, some categorical dimensions consolidated into index form.

3) Standardised characteristic score modifiers to separately apply weightings, further reduction of dimensions, collapsing some categoricals further into indices for ease of application - at possible expense of increased subjectivity.

4.2 Placing Assets in TS10

Having produced these proto-taxonomies, the Nickerson method was applied and selected major cryptographic assets were “classified” with meta-characteristic scores for Securityness ($\mathcal{{\overline{S}}}$), Moneyness ($\mathcal{{\overline{M}}}$) and Commodityness ($\mathcal{{\overline{C}}}$) in order to populate this TS10 instantiation of TokenSpace. Assets were selected by their perceived importance via the admittedly coarse heuristic of market capitalisation at time of writing (Q1 2019).

Assets included: Bitcoin (BTC), Ethereum (ETH), Ripple (XRP), EOS (EOS), Litecoin (LTC), Bitcoin ABC (BCH), Tether (USDT), Stellar (XLM), Tron (TRX) and Binance Coin (BNB).

TS10 full breakdowns of scores and weightings for each meta-characteristic's taxonomy are presented in Tables 12, 13 and 14 (present in this PDF). Overall score values are presented in Table 15 and definitions of dimensions in Table 16. Statistical analytical results are visually represented in Figures 18 & 19, and the TS10 TokenSpace is visually represented in Figure 20.

DISCLAIMER: A reminder that the decisions of category selection, dimension weighting and / or index values have been made in an ad hoc, approximate and subjective manner and do not necessarily correlate to an objective representation of reality. The author is not a lawyer, regulator or legal professional and has no definitive opinion on the regulatory or compliance status or consequences of assets being classified with particular assignations by any territorial or jurisdictional legislature. By reading this document you agree that the author accepts no liability or responsibility for the results outlined below or any discussions arising thereof. TS10, TSL7 and TSTDX TokenSpace scores are provided for intellectual purposes and the aforementioned TokenSpaces are an abstract and hypothetical representation based upon the methodologies developed in this work.

Considering first the scores realised by the TS10 TokenSpace in Table 15, it is clear that there is significant variation between the values obtained for Securityness ($\mathcal{{\overline{S}}}$), Moneyness ($\mathcal{{\overline{M}}}$) and Commodityness ($\mathcal{{\overline{C}}}$) meta-characteristics across the 10 cryptoassets with the highest nominal market capitalisations.

Securityness in TS10 exhibits a minimum of 0.088 for BTC, a maximum of 0.81 for BNB, a mean of 0.55 and standard deviation of 0.27. The range of values is wide at 0.72, as some cryptoassets possess very high values whilst others possess low or moderate values. There is also considerable skew in the distribution of these values, as the maximum value is 0.96 standard deviations from the mean whilst the minimum value is 1.7 standard deviations away.

Moneyness in TS10 exhibits a minimum of 0.06 for BNB, a maximum of 0.41 for BTC, a mean of 0.16 and standard deviation of 0.11. The range of values is comparatively compact at 0.35, a consequence of all cryptoassets possessing modest or low scores. Significant skew is also present in the distribution of these values, as the maximum value is 2.3 standard deviations from the mean whilst the minimum value is 0.91 standard deviations away.

Commodityness in TS10 exhibits a minimum of 0.12 for EOS and BNB, a maximum of 0.68 for BTC, a mean of 0.28 and standard deviation of 0.17. The range of values is significant at 0.56. There is also considerable skew in the distribution of these values, as the maximum value is 2.35 standard deviations from the mean whilst the minimum value is 0.94 standard deviations away.

Figures 18 and 19 depict histograms and boxplots for each of the three meta-characteristics in linear and logarithmic scales to visually display the distibution of scores. Interestingly the statistical computing software employed here (R) initially treated the $\mathcal{{\overline{M}}}$ and $\mathcal{{\overline{C}}}$ values of BTC as outliers due to their significant difference to the other cryptoassets considered. In the author's opinion this is a reflection of Bitcoin's unique status as a leaderless and permissionless pseudo-monetary commodified good with a paucity of attributes in common with other cryptoassets. In other words, Bitcoin is sui generis among other cryptoassets, which themselves are sui generis to lesser and varying degrees in comparison to legacy assets.

Figure 20 contains a three-dimensional view of the TS10 TokenSpace with included assets occupying the coordinates arising from their scores. As intended in the design of the TokenSpace methodology, this affords ready visual comparison of asset locations. From this visual representation, it becomes more readily apparent that there are several sub-populations of cryptoassets within TS10. BTC occupies a domain of its own, as do ETH and USDT to a lesser extent. This is unsurprising as these three cryptoassets are significantly differentiated: Bitcoin as a highly decentralised P2P commodity money, Ethereum as a scripting platform and Tether as a collateralised stablecoin. Ethereum's Securityness largely arises from its initial issuance mechanism - its token crowdfunding was the largest “ICO” at time of launch - and the presence of a powerful foundation and leadership class who largely influence and fund the course of action of the network [123].

BCH and LTC occupy locations close to each other, as minor analogues of Bitcoin they exhibit some similarities to BTC but have much weaker value propositions as monetary or commodity goods due to inferior network security and the presence of single points of failure such as prominent leadership. However like Bitcoin they did not issue assets via a token sale and solely rely on proof-of-work, keeping their Securityness fairly low.

XRP and XLM appear close together which is unsurprising as they share many of the same characteristics, with XLM originating as a codebase fork of XRP by the same progenitor (Jed McCaleb), with both networks having a heavy concentration of asset supply residing with insiders and / or foundations. Network nodes of both XRP and XLM are challenging to run permissionlessly with high concentrations of validating nodes in the respective federations being controlled by Ripple Labs and IBM / Stellar Foundation respectively [124, 125]. XRP transactions have been prevented from occurring due to disputes between network controllers and estranged insiders, and a major historical covert supply inflation event was recently uncovered in XLM which allowed an attacker to create several billion tokens [126]. For these reasons and others, XRP and XLM are poor monetary or commodity assets but do display a fairly high degree of Securityness.

EOS, TRX and BNB are the final subset of cryptoassets in TS10. All three were intially issued via ICOs, and exhibit a high degree of centralisation in monetary, network, architecture and / or stakeholder influence with limited asset and / or network utility at time of analysis in early 2019. These cryptoasets exhibit high Securityness, low Moneyness and low Commodityness, making them possible targets for regulators looking for high profile cases to investigate. It is somewhat apparent from the operations of these projects that this has been considered to be a risk, with EOS creator Block.One locating themselves in the British Virgin Islands and offering non-functional tokens in their ICO with no promise to launch a network. Binance Coin is issued by the sprawling exchange group Binance, which is commencing exchange operations in new jurisdictions faster than regulators can react having nominally relocated to Malta last year. The token burning and exchange fee discounts for token-holders give BNB a very high degree of likeness to a classical securitised asset. Tron is a project which seems to be mostly focused on marketing with a constant stream of giveaways and partnership announcements, a rather consistent record of questionable veracity of claims, initially uncredited re-use of code (EthereumJ), whitepaper (IPFS / Filecoin), dubious claims and explicit promotion of the speculative potential of the asset [127].

4.3 Cluster Analysis & Correlations

Two statistical analysis methods were employed to further understand the anisotropy of the asset locations in TS10, k-means and agglomerative hierarchical clustering [128, 129]. Each approaches the dataset in different ways to reach a set of ending conditions, in a manner not unlike taxonomies themselves. The k-means algorithm creates k number of cluster centroids (with k being adjustable or optimisable) before iteratively assigning values to the closest centroid and adjusting the updated position of the centroid until no further changes take places between iterations. One key assumption made by the algorithm is that data is isotropic (or spherical), which may render it ineffective for advanced TokenSpace studies employing higher dimensionalities and / or anisotropic PDFs as discussed in Section 3.3.8.

Agglomerative hierarchical clustering in contrast incrementally builds clusters, producting a dendrogram. The algorithm first assigns each sample to its own cluster, with each step incorporating a merge of the two most similar clusters until all have been merged. When perceived in reverse, the dendrogram resembles a stepwise sorting machine, sub-dividing the population of objects on the basis of similarity. No centroid parameter is required.

For k-means clustering of TS10, it was found that values of k of 3 or above produced acceptable levels of extrinsic variation between the clusters versus intrinsic variation within the sum of squares of clusters (Figure 21). To correlate with the visually-derived groupings above, 6 clusters were judged to be optimal but the analysis could also be conducted using 4 or 5. Results from k-means clustering with values of k = 3, 4, 5 and 6 are displayed in Figure 22.

Hierarchical complete-link clustering was conducted with the dendrogram produced exhibited in Figure 23. Complete-link refers to the clustering method, whereby in each iteration, merging occurs between two clusters which possess the smallest maximum pairwise distance. In contrast single-link clustering merges the two clusters with the smallest minimum pairwise distance. In this case, both techniques were employed and the complete-link approach resulted in the greater similarity between clusters as measured by agglomerative coefficient, with 0.84 versus 0.77 for single-link [130]. As discussed above, when the clustering process is considered in reverse, it takes on some of the properties of a “sorting machine”, allowing the degrees of similarities and differences between assets to be readily parsed. Largely corroborating previous results, the process found BTC to be unique among the TS10 basket of cryptoassets, with further clusters ETH and USDT, LTC and BCH, XRP and XLM and EOS, BNB and TRX. The principal difference in results between this approach and the 6 cluster k-means is that ETH and USDT are not differentiated here, making it more similar to the 5 cluster k-means results.

Correlations between the three meta-characteristics were also explored pairwise and are listed in Table 17. Pearson's correlation coefficient (Rho) can take values between positive one and negative one, representing perfect positive and negative correlation of two datasets respectively. As would be expected from the prior findings, in TS10 there is a high degree of positive correlation between Moneyness and Commodityness with a Pearson's Rho of 0.99, indicating that indeed Moneyness and Commodityness are highly related attributes - based upon the empirical data and subjective inputs to TS10 at least. Correlation of Securityness to Moneyness and Commodityness are similar and strongly negative with values of -0.88 and -0.87 which is to be expected for meta-characteristics which required dimensions to be inverted when used across taxonomies.

4.4 TSL7: Using TokenSpace to Compare Cryptographic & Legacy Assets

A separate TokenSpace construction to TS10 will now be presented: TSL7, with the goal of exemplifying the ability of TokenSpace to compare and contrast legacy assets alongside cryptoassets. No taxonomies have been developed for this TokenSpace, instead relying on the TS10 data for cryptoassets and intuitively reasoned scores for legacy assets. Tables for overall scores and each meta-characteristic are shown in Tables 18, 19, 20 and 21 with visual representation in Figure 24.

Apple stock (AAPL) represents an ideal type of a securitised asset with negligible monetary or commodity attributes, with soy beans (SOY) and gold metal (GOLD) being canonical examples of consumable and non-consumable commodities respectively. The US Dollar (USD) still possesses a strong Moneyness being the de facto world reserve currency, though as is seen in Section 4.5 its Moneyness and Commodityness have been falling due to governmental and central bank policy choices. One interesting observation is the near-equivalence in TSL7 of the Moneyness of gold metal and Bitcoin (BTC). There appears to be a reversal of the perceived premier commodity monetary good as human society continues to engage in technological and particularly digital advances.

The “Lindy type time-dependence” indexed dimension values were considered to be universally low for all cryptographic assets due to the lack of ecosystem maturity, with a “Lindy index” of 1 for gold, which has been considered valuable by humans for no less than several thousand years [131].

4.5 TSTDX: Time-Dependence of Selected Assets

A third TokenSpace construction will now be presented. TSTDX has been constructed with the goal of exemplifying the ability of TokenSpace to map time-dependence of asset attributes. As with TSL7 no taxonomies have been developed for this TokenSpace, instead relying on the TS10 data for cryptoassets in 2019 and intuitively reasoned scores for present day scores of legacy assets and all historical values. Tabulation of overall meta-characteristic scores are shown in Table 22 with visual representation in Figures 25, 26, 27, 28 and 29. Intuitive judgement was applied to give an indicative depiction of how a time-dependence TokenSpace such as TSTDX may be more rigorously constructed in future.

A number of interesting observations can be made in TSTDX. It is apparent that monetary metals such as gold (GOLD) and silver (SILVER) are decreasing in Moneyness with time as Bitcoin's (BTC) increases - ostensibly as the digitalisation of human society corresponds to favouring similarly digital (“simulacrised”) money such as Bitcoin over specie.

The loss of gold and silver backing on moneys such as the British Pound (GBP) and the US Dollar (USD) leading to loss of Commodityness, Moneyness and an increase in Securityness may also be rationalised as derealisation - a loss of mimetic gravitas in addition to simulacrum-related societal sentiment. Related is the loss of Moneyness of gold and silver over time, as these two metals have long been on a path to demonetisation. In this respect, silver is some way ahead of gold, being largely a commodity rather than a commodity-money in the present day.

The time-dependence of cryptographic assets generally shows a trend of decreasing Securityness as the networks mature and assets become more adopted, distributed, widely held, useful and used. In concert Moneyness and Commodityness also tend to increase as more reasons to use, hold and transact with the assets emerge. Ethereum (ETH) is particularly remarkable as - in tandem with Hinman's summer 2018 sentiments as discussed in Section 1.3.3 - what started as a securities offering of a centralised asset reliant on the efforts of a team of others for speculative gain has become (to some extent) more widely used, useful, held and distributed hence leading to a decrease in Securityness and increases in Moneyness and Commodityness. It could perhaps be said that Ethereum in particular is well on the path to desecuritisation, or indeed may have arrived at that destination depending on where boundaries are perceived to lie in TokenSpace. The US Dollar (USD) still possesses a strong Moneyness being the de facto world reserve currency, though as is seen in Section 4.5 its Moneyness and Commodityness have been declining since the abandonment of gold-backing and the rise of the petrodollar system.