Nassim Taleb coined the term “anti-fragile” to describe systems that benefit from shocks. Crypto economic systems may display fragile, robust, and anti-fragile characteristics.
By Percy Venegas www.EconomyMonitor.com
Summary
Value in algorithmic currencies resides literally in the information content of the calculations; but given the constraints of consensus (security drivers) and the necessity for network effects (economic drivers), the definition of value extends to the multilayered structure of the network itself — that is, to the information content of the topology of the nodes in the blockchain network, and, on the complexity of the economic activity in the peripheral networks of the web, mobile apps, mesh-IoT networks, and so on. In this boundary between the information flows of the native network that serves as the substrate to the blockchain, and that of the real-world data, is where a new “fragility vector” emerges. Our research question is whether factors related to market structure and design, transaction and timing cost, price formation and price discovery, information and disclosure, and market maker and investor behavior, are quantifiable to the degree that can be used to price risk in digital asset markets. We use an adaptive artificial intelligence method to study the adaptive system of crypto currency markets. The results obtained show that while in the popular discourse blockchains are considered robust and cryptocurrencies anti-fragile, the cryptocurrency markets may display fragile features.
Intro
There have been numerous cases documented of malfunctions in the cryptocurrency markets; some have been related to the internal operation of exchanges, such as the glitch in the cryptocurrency exchange GDAX that crashed the price of Ether from $319 to 10 cents (Kharpal); others have to do with the dissemination of information, such as the negative impact on bitcoin prices after the de-listing of Korean exchanges in Coinmarketcap, a popular price tracker (Morris).
While the bitcoin network is decentralized by design, the web and other peripheral networks where the services that enable the bitcoin economy operate are not. This extends to any other cryptocurrencies with a sufficiently high degree of decentralization, such as Ether. Roll (ROLL 1984) demonstrated that volatility is affected by market microstructure, and by applying this idea to the off-chain side of a crypto-economy we are able to develop real-life risk metrics for the blockchain financial system.
Identifying systemically important services in cryptocurrency markets
At the time of writing, March 6th 2018, 3:38 PM (UTC), the total market capitalization of the cryptocurrency markets (including coins, excluding tokens) was $405,939,039,748. The largest assets by capitalization are bitcoin (BTC) $187,065,686,782, and ether (ETH) $81,602,911,777.
To begin the development of an analytical model which is behavioral finance-aware, we should consider that in the retail segment, people usually utilize one or more public services before checking the status of, or performing a transaction. Those services include price trackers targeted at different regional markets, trading platforms of various types, centralized exchanges and instant exchanges, wallets, among many others. We use a common usage metric; a sample of the largest 1000 contributors to the Bitcoin network and 1000 contributors to the Ethereum network (measured at points of consumption) are shown in Figure 1.
Figure 1. BTC and ETH off-chain supporting services
Given the different use case of each crypto asset, each tends to have a number of specialized services around its orbit. However, many others are shared (with a higher density at the BTC side, top left). If we start mapping the relationships between services as well, the problem will quickly become intractable.
The common sources in both sets bring the variables of interest from 1000 to 196, so we can further prioritize sources by traffic contribution. But ultimately we use an adaptive approach as a way to detect invariance, and capture general relationships by using prices are the response variables. This treatment of the data reduces the variable set to 22 (see Figure 2). The adaptive approach produces models that compete to reduce the error while maintaining the lowest degree of complexity. In the accuracy-complexity Pareto of the figure, optimal models (with the best possible trade-off) are depicted by red points. We also note, by the variable presence across models, that only a handful of sources are of material importance.
Figure 2. Variable Map and Pareto Front
Systemic risk mapping
It is possible to study multivariate volatility using traditional methods, but let us take advantage of the expressions generated to construct a graph representation of system dependencies.
Detection of (Anti) Fragility
Formally, fragility and antifragility are defined as negative or positive sensitivity to a semi-measure of dispersion and volatility (Taleb 2013). What is different in this case is that we are not concerned with exposure to price shocks only, but prices themselves can change abruptly in response to events from the environment. By constructing a graph using formulas selected from our adaptive exploration of the model space we obtain a network as the one in Figure 3.
Figure 3. Network of risk relationships
The network reveals the sources of fragility, overlapping exposures, and, relevant feedback effects. For instance, it becomes clear than Ether prices are affected by changes in BTC prices and not the other way around. It also can be understood what are the shared risk drivers (e.g. the usage of X1 has an effect on both BTC and ETH), which allows for more control over aggregate factor exposure (in hedge fund speak).
This approach also allows uncovering secondary effects. A bet on X11 and X3 may prove ineffective since those have a relationship of mutual dependence — but their risk profiles are different since X3 is exposed to both crypto asset prices directly; they also affect the environment differently — X3 has a higher outdegree than X11 (more outgoing connections). In this particular case this makes sense: in practice, a wallet service will be more affected by activity in selected exchanges, than directly by prices.
Finally, the method makes tractable the nodes of systemic importance, e.g. X1, X3, and X13, all have a relationship with bitcoin prices. This is something that would be impossible to find in a traditional network representation such as Figure 1.
Regulatory and Cyber risk
The risk network also offers an opportunity to disambiguate opaque situations. For instance, X13 appears to portrait a spurious relationship (how can a service affect the price of a decentralized currency?), but a closer inspection reveals the occurrence of a extreme/tail event. After the crackdown of the Chinese government on bitcoin exchanges in September of 2017, one of the most popular exchanges rebranded and initiated operations as an international service (but still, being accessed mainly from mainland China). As we can see in Figure 6, the remarkable growth in what would be an impossible timeframe for most online services (several orders of magnitude in less than 6 months) not only explains the behavior captured by the model, but provides evidence of the increase on the survival rate of systemically important nodes in the crypto economy. Some services prove hard to kill, even if they operate in the centralized part of the economy. In this sense, those components enhance the anti-fragile characteristics — they simultaneously boost the underlying, and may even benefit from the shocks.
Unique users X13 exchange
But despite resilience in the structure of the network, attack vectors and vulnerabilities in the technology used create conditions for fragility in the system. For instance, we found that price trackers of systemic importance use content delivery services such as Amazon S3. On February 28th, 2017 an outage on S3 buckets in northern Virginia, US, made Amazon Web Services inaccessible. Software, from web apps to smartphone applications, relying on this cloud-based storage broke, taking out a sizable part of the internet (Nichols). Bitcoin reached its current price levels after rallies in May-June 2017 and December 2017, if a catastrophic event such as the AWS outage of early 2017 occurs now that bitcoin has more broad adoption among consumers and is more intertwined with the mainstream financial system, price shocks are likely.
Discussion
Cryptocurrency markets appear to showcase the malaise of boom, bust, and failures to learn common in experimental markets (Paich 1993):
Word of mouth, marketing, and learning curve effects can fuel rapid growth, often leading to overcapacity, price war, and bankruptcy. Previous experiments suggest such dysfunctional behavior can be caused by systematic ‘misperceptions of feedback’, where decision makers do not adequately account for critical feedbacks, time delays, and nonlinearities which condition system dynamics. However, prior studies often failed to vary the strength of these feedbacks as treatments, omitted market processes, and failed to allow for learning.
Not only systemically important services rely on third-party providers that may have exposures, some are inherently fragile. We found that one of the most important information services in the economy is a forum built using open source technology that is known for scalability concerns and constant security exploits; the forum operators probably never planned to reach such a large scale, so quickly. In this sense, cryptocurrency markets may be suffering from the Getting Big Too Fast problem (Sterman 2007), when market dynamics are rapid relative to capacity adjustment.
But more importantly, the cryptocurrency markets show fragility when uncertainty increases the probability of dipping below the robustness level supposedly conferred by the underlying technology of the digital assets (blockchain).
Here is where an adaptive approach to study the adaptive system of the cryptocurrency markets becomes instrumental to the identification and mapping the sources of risk — whether this is psychologically bound to trust-asymmetries, or fundamental in nature, as expressed by the progression from conditions of anti-fragility to robustness to fragility. Interestingly, the creation of new methodologies to understand the crypto economy, given the shortcomings of traditional economic theory, may have an unexpected effect: the advent of decentralized ledgers may also accelerate the transition from the economic equilibrium paradigm to the realm of economic complexity (Bheemaiah 2017 Kary Bheemaiah).
As discussed previously, the effects work in both ways, the deviations in service usage can also affect price. We must also acknowledge that for those services hosted in centralized networks there are defacto equivalents to “option values” abscribed due to operational activities — either because it has been assigned for internal budgetary purposes, or because attention is actually traded in public markets such as two-sided ad marketplaces. But since the ultimate purpose of migrating to blockchain technology-based digital assets is to tokenize value, we can conceive that in the near term options priced as a multiple of a certain engagement metric will be measurable inputs to attention-based risk frameworks. And given the theoretically unlimited number of assets that can be created, the adaptive analysis approach should remain observant of Occam’s razor principle: among the theories that are consistent with the observed phenomena, one should select the simplest theory (Li 2008). Ultimately, the goal is to enable prediction under above-average degrees of uncertainty, but with attainable resources — the optimal trade-off of accuracy and complexity.
Conclusions
At the present time, cryptocurrency markets display fragile characteristics. But one should be encouraged with the anti-fragility characteristics discovered in the system, albeit if these are defined in broad terms. The open problem remaining is that fragility is K-specific, i.e. we are only concerned with adverse events below a certain pre-specified level, the breaking point. The multipe exposures, of materially different natures, make the task of defining that level more daunting. As an example, think on the case of specialized social networks. Given the complexity space analyzed, the high contributors operating an hybrid model of blog platform and social network were not discussed (think Medium). Some of those services are instrumental on shaping public opinion and influencing investor behavior, however the relevant signals are “tacit” knowledge embeded in the network. Here is where the formulation of economic complexity indexes, with their treatment of ubiquity of services and divertisity of economies, serves as a foundation to characterize and quantify drivers of growth. Enhancing this methodological foundation with an adaptive factors-based analytical approach provides the tools to understand fragility, and appropiately price risk, in the cryptocurrency markets.
Published by www.EconomyMonitor.com
Information provided for educational purposes only, should not be construed as financial or legal advice.
Recommended reads Crypto economy complexity