https://orcid.org/0000-0003-4857-0195
Figures
Abstract
In contrast with robust systems that resist noise or fragile systems that break with noise, antifragility is defined as a property of complex systems that benefit from noise or disorder. Here we define and test a simple measure of antifragility for complex dynamical systems. In this work we use our antifragility measure to analyze real data from return prices in the stock and cryptocurrency markets. Our definition of antifragility is the product of the return price and a perturbation. We explore different types of perturbations that typically arise from within the system. Our results suggest that for both the stock market and the cryptocurrency market, the tendency among the ‘top performers’ is to be robust rather than antifragile. It would be important to explore other possible definitions of antifragility to understand its role in financial markets and in complex dynamical systems in general.
Citation: Alatorre D, Gershenson C, Mateos JL (2023) Stocks and cryptocurrencies: Antifragile or robust? A novel antifragility measure of the stock and cryptocurrency markets. PLoS ONE 18(3): e0280487. https://doi.org/10.1371/journal.pone.0280487
Editor: Sebastián Gonçalves, Universidade Federal do Rio Grande do Sul, BRAZIL
Received: May 2, 2022; Accepted: December 29, 2022; Published: March 16, 2023
Copyright: © 2023 Alatorre et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: We analyzed a data set with historical daily prices and volumes of more than 7000 stocks from the US Stock Market (https://www.kaggle.com/datasets/borismarjanovic/price-volume-data-for-all-us-stocks-etfs). We truncated the data starting from 1990. This data set considered only active stocks by the end of 2017 and there is no data from stocks that disappeared from the market before that date. In the case of cryptocurrencies, the data (https://www.kaggle.com/datasets/jessevent/all-crypto-currencies) consisted of historical daily data (open price, market capitalization and volume) of around 1800 assets, the oldest being from 2013 and up to November 2018. Instructions for accessing the processed data sets can be found in this purl: https://purl.prod.archive.org/purl/AFSCM_data/processed-data.
Funding: CG - Award from Fundación Marcos Moshinsky A. C. (http://www.fmm.fisica.unam.mx/) -UNAM PAPIIT Programa de Apoyo a Proyectos de Investigación e Innovación Tecnológica (https://dgapa.unam.mx/index.php/impulso-a-la-investigacion/papiit): IN107919 and IV100120 grants. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
1. Introduction
Some systems have the property of being robust, in the sense that they are not affected by the presence of noise. An analogous concept of robustness is stability, where a complex dynamical system recovers its original stable state after a small perturbation. On the other extreme, we have the property of being fragile, where the presence of noise breaks or collapses the system. The case of fragility corresponds to a state in unstable equilibrium, where a perturbation drives the system away from the original state.
Now, let us describe the counterintuitive concept of antifragility, which is different from robustness and fragility. A system is antifragile when it benefits from noise or disorder. Thus, noise acts as a constructive element that allows the system to perform a function successfully. Therefore, one expects to encounter antifragility in a complex dynamical system that is nonlinear, has many degrees of freedom and is out of equilibrium.
The term antifragility was coined by Taleb in his book of the same name [1]. Taleb’s ideas have been further explored in different contexts not only in risk analysis and financial systems [2–4], but also as a means for strategic design and planning [5–7]. The original formal definition [8] (see also [9, 10]), however, is difficult to use in practice and with real data. Therefore, we use in this paper an alternative definition of antifragility that can be implemented easily for a quantitative analysis and is general enough to be applied in a broad range of domains.
This definition was inspired by previous recent work, where it was proposed a pragmatic antifragility measure for complex systems, provided one can define measures of “satisfaction” for each of its agents and “perturbation” for the whole system. These measures are not supposed to be dependent on each other, as in Taleb’s original idea, which assumes that the satisfaction is a function that depends on the magnitude of the perturbation. In this work, antifragility is defined as the product of perturbation and satisfaction. Consequently, an agent is robust if its antifragility is close to 0 and fragile if it is negative. The agent is antifragile for positive values of the measure. This new definition of antifragility has been already applied with promising results in the context of random and biological Boolean networks [11], multilayer random Boolean networks [12], convolutional neural networks [13], and ecosystems [14].
In this work we tested this antifragility measure with real data from the stock market and the cryptocurrency market. In this context of economics and finance, we used the price returns of assets as a measure of satisfaction, while a perturbation measure for each system was defined as the mean of different volatility parameters coming from within the markets–involving the price returns, the market capitalization and the volume (amount) of transactions of each asset, as well as a global mean index and the volatility index VIX (the stock symbol and the popular name for the Chicago Board Options Exchange’s CBOE Volatility Index, a popular measure of the stock market’s expectation of volatility based on S&P 500 index options). The antifragility measure was calculated as the price return multiplied by the perturbation. We evaluated this antifragility measure in daily, weekly, monthly and semiannual time-scales, and compared it with different “good-performance” measures that we could extract from the data.
The identification of antifragility could be helpful in a wide range of scenarios. Particularly in economic and financial systems, which Taleb [1] suggested that should be antifragile. This work aims to apply Taleb’s ideas to real data of the stock and cryptocurrencies market.
The paper is organized as follows: The next section gives some background of other physical systems that “gain from disorder”, and the role of randomness and noise in financial systems. Then, we explain our definition of antifragility in general and in the context of stock and cryptocurrency markets. In the following sections, we describe the data and the variables employed and the results. Finally, we end with a discussion and concluding remarks.
2. Background
In the introduction of this paper, we mentioned that antifragile systems gain in the presence of noise. This noise can come from an external perturbation in an open system or can be an internal noise due to the complexity of the system itself. However, this idea has been known in the field of complex dynamical systems, nonlinear stochastic dynamics and statistical physics for more than 40 years. Thus, the concept of antifragility can be traced back to the studies of the constructive role of noise in these nonlinear systems out of equilibrium. We will very briefly describe some of these phenomena that have been thoroughly studied in the literature: stochastic resonance, noise-induced transport and optimization by simulated annealing. All these phenomena share the property of gaining in the presence of a stochastic signal or perturbation, thermal noise or by being exposed to fluctuations.
2.1 Constructive role of noise in the physical sciences
2.1.1 Stochastic resonance.
The idea of stochastic resonance is to add noise to a signal and measure the signal to noise ratio (SNR). Contrary to what is observed in linear systems, where the addition of noise decreases the SNR, some nonlinear systems exhibit an increase of the SNR as the intensity of the noise is increased; the signal reaches a maximum after which the SNR starts to decrease. So, a SNR is a function of the intensity of the noise and has a maximum, similar to a resonance in a physical system; thus, the name of stochastic resonance. It is worth mentioning that the phenomenon of stochastic resonance is manifested when the system is driven out of equilibrium by applying an external force. Thus, even though the original system can be in equilibrium without any external force or perturbation, once we perturb the system, the full system (the original one plus the external forcing) can be considered as an out-of-equilibrium system. This phenomenon can be observed in physical and biological systems alike. The idea was introduced in the early 1980s by G. Parisi and collaborators. For a thorough review see [15]. Therefore, adding noise is beneficial and, in this sense, is closely related with the idea of antifragility.
2.1.2 Noise-induced transport and Brownian motors.
These phenomena refer to the counterintuitive idea of incorporating noise or a stochastic signal to enhance and generate transport in an asymmetric physical system. The asymmetry can be spatial or temporal and it rectifies a symmetric noise generating a finite current. The phenomenon was studied by Richard Feynman in the early 1960s and gained attention in the 1990s with the study of thermal ratchets and Brownian motors that use Brownian motion on top of asymmetric ratchet potentials. One of the key motivations was to understand the mechanism behind motor proteins inside the cell and applications to nanoscience. For a review, see [16]. For a combination of Brownian motors and stochastic resonance, see [17]. Again, the main idea is that adding noise can be beneficial, as in an antifragile system.
2.1.3 Optimization by simulated annealing.
This is an important phenomenon where one implements simulated annealing, as in statistical mechanics, using thermal noise to escape from local minima in a landscape to arrive at a global minimum; thereby finding an optimization function. There is a vast literature around this idea. See for instance the original paper [18] and also [19].
There are other phenomena related to the idea of antifragility, like noise-induced synchronization in non-linear physical and biological systems [20], the “order from noise” [21], and the “complexity from noise” [22] principles.
2.2 Noise and fluctuations in economics and finance
Other important examples of complex dynamical systems that might be antifragile are economic and financial systems. Clearly, both are complex systems that are open, nonlinear, and have many agents that operate collectively, giving rise to emergent properties like the ones we observe in financial markets and other macroeconomic indicators. The question is if we can find the phenomenon of antifragility in economics and finance. Recent developments have used different definitions of antifragility to study recovery [23], collaborative business ecosystems [24] and the purchasing and supply chain management [25], but all of them go back to Taleb’s original definition. Our approach is more heuristic in the sense that it aims at the identification and measurement of antifragility.
2.2.1 Noise.
Let us start with noise. In the context of economics and financial markets, the idea of noise can have different meanings. Noise can emerge as a result of a large number of small events due to the collective actions of economic agents on a daily basis. On the other hand, a small number of extreme events, like crashes, can have severe consequences as well. Noise makes trading in financial markets possible, and thus allows us to observe prices for financial assets. Noise causes markets to be somewhat inefficient, but often prevents us from taking advantage of inefficiencies. Noise in the form of uncertainty about the actions to be taken by a large number of agents in the future makes it very difficult for us to adopt rational decisions, and act by guessing or intuition instead. As a result, we tend to consider price fluctuations and price return in financial markets as a random or stochastic process, and act mainly by guessing. We are forced to act largely in the dark. See the classical paper of 1986 by Fisher Black titled simply “Noise” [26]. The topic of noise has a vast literature and extends in many directions, from mathematics and natural sciences, all the way to medicine, psychology, economics, politics, and so on. See the very recent book by Daniel Kahneman titled “Noise: A flaw in human judgment” [27].
2.2.2 Perturbations.
Regarding the perturbations that we use in this paper, we consider only internal perturbations (or noise) that comes from inside the market itself. This endogenous noise of the markets include the fluctuations observed in the returns of prices, volatility and market capitalization. As mentioned before, this noise emerges from the collective activities of many economic agents. Many of these agents do not have information and sometimes are called “noise traders” after the classical description by Black [26]. See also the book by Sornette [28], where different models are analyzed. These noise traders, among other internal considerations due to the complexity of markets, generate smaller scale shocks. However, financial markets can be affected by aggregate shocks as well; see [29, 30]. These aggregate shocks arise due to external or exogenous events to the market, such as wars, terrorist attacks, massive migrations, political turmoils, or a pandemic. In our study here, we are not considering the effect of these aggregate shocks, but it would be important to take them into account. That would be an interesting extension to consider for future research.
2.2.3 Price returns.
Finally, let us discuss the last piece needed to construct our antifragility measure. We will see that this measure is the product of price returns and perturbations.
Price returns are one of the central quantities to study in the financial markets. It refers to the difference in prices of an asset evaluated at two different times. It can be a time difference of seconds, days, weeks, years or decades. There are several definitions of these price returns that can be ratios or percentage of price fluctuations, or differences between the logarithm of price (the so-called log returns). This central quantity is considered to be a random variable, according to the discussion above, and is therefore natural to study its statistical properties. In particular, it is important to study the probability distribution of price returns. An important result that has been established is that the stock returns probability distribution has a power law or fat tail characterized by an exponent close to three, the so-called cubic law (see [31] for a recent review of this cubic law and other power laws). These power-law tails in the distribution have important consequences in the dynamics of the price returns: due to these long tails, we expect to have large fluctuations and extreme events in the time series. These extreme events correspond to crashes in the stock market and, from the viewpoint of the statistical physics of complex systems, can be understood as a scale-free dynamics near criticality [28]. This scale-free dynamics implies that the system is prone to have large fluctuations and, in this sense, becomes less stable. The consequences of the power-law distributions of the returns have triggered a whole new field of study called econophysics [32, 33], where there is an emphasis in trying to understand the origin of this long-range fat-tails distributions. The main idea is to model economics and finance as a complex system using the tools of statistical physics and nonlinear dynamics: for instance, using agent-based models as a “microscopic” description to explain the emergent phenomena observed in financial markets. At the end, the scale-free power laws observed in the return distribution implies some sort of phase transition where the system can become unstable (or metastable) locally and starts to develop a long-range dynamics.
2.2.4 Complexity in economics and finance.
It is worth mentioning briefly new developments of complexity science in economics and finance. There is a vast literature around the topic. Recent papers dealing with the collective dynamics of stock markets, universal features and complexity in financial data are [34–37] and, in particular for the analysis of cryptocurrency markets [38–43].
Finally, the application of the emergent science of networks to financial systems is an important field of research; for a very recent review see [44] and references therein. Other important reviews on the foundations of complexity economics are [45] and in economic complexity theory and applications [46]. See also the book by Hidalgo [47].
It would be interesting to consider for future research the incorporation of the concept of antifragility as a feature of complex dynamical systems in general, and in particular from the viewpoint of complexity and networks in economics and finance.
3. Materials and methods
We now describe the data set that we used throughout this paper.
We analyzed a data set [48] with historical daily data (open price and volume) of about 7,000 stocks from the US Stock Market. Such a dataset considered only active stocks by the end of 2017 and there is no data from stocks that disappeared from the market before that date. We truncated the data starting from 1990.
In the case of cryptocurrencies, the data [49] consist of daily historical data (open price, market capitalization and volume) of around 1800 assets, the oldest one starts in 2013 and ends in November 2018.
The raw data may be found in [48, 49]. And all the processed data is available in this purl: https://purl.prod.archive.org/purl/AFSCM_data/processed-data.
3.1 A measure of satisfaction
We will start our analysis with the variation in time of the price of a stock or cryptocurrency. Let pi denote the price of a stock, and qi indicate the price of a cryptocurrency, at time i, where i = 1,2,…,N is a discrete variable indicating the time when the price is measured. The stock return is the difference between the prices of two consecutive times. That is:
(1)
This quantity, the stock return, can be considered as a measure of “satisfaction”, being positive when the price increases and negative otherwise. We will refer sometimes to the stock return in Eq (1) as “satisfaction” in order to connect with a set of previous papers of one of the authors (CG), in which the concept of satisfaction was used–together with this definition of antifragility–for different fields such as ecology and networks (see [11–13]).
3.2 Definition of antifragility
In order to quantify how sensible the stock return of a given asset is to an external perturbation, we introduce here a definition of antifragility. We consider markets with stock returns S and an external time-dependent perturbation P. Let S(x,i) be the stock return of asset x at time i, and P(i) be the perturbation measure of the system at time i. We define the antifragility of asset x at time i as:
(2)
The global antifragility of asset x is then defined as the mean value of A(x,i) over the whole time interval under consideration, i.e.
(3)
where the ⟨⟩i denotes to the arithmetic mean of the set over index i
(4)
This definition of the concept of antifragility allows us to classify the systems as antifragile, robust and fragile depending on the overall result of the stock returns and the perturbation. We consider that a stock is antifragile if the stock return is positive even in the presence of a perturbation. On the other hand, a stock is robust if the stock return is not affected by the perturbation. Finally, a stock is fragile if the stock return decreases in the presence of the perturbation. These considerations apply both to stocks and cryptocurrencies.
3.3 Normalization for returns and perturbations of stocks and cryptocurrencies
Throughout this study two different systems were considered: the stock market and the cryptocurrency market. We used in each case different perturbation measures based on the available data. However, we consider the return (price differences) as a measure of satisfaction in both cases.
Since we are interested in measuring and comparing the antifragility in different systems, we normalized the stock return S(i) and the perturbation P(i) in such a way that both quantities are bounded in the intervals [–1, 1] and [0,1], respectively.
For the stock return S, we use a new normalized return S’ as
(5)
where |S| is the absolute value of S, and max(|S|) is its maximum. With the definition given by Eq (5), S’ is a real number in the interval [–1,1].
For the perturbation P, we use a new normalized perturbation P’ as
(6)
Clearly, P’ is a real number in the interval [0,1].
These normalizations imply that the antifragility A’ = S’·P’, is a real number in the interval [–1,1].
In the following sections, we will assume that all the perturbations and returns have been already normalized in this way. We will also assume that the quantities under consideration such as the price pi, the volume vi (the number of shares that are sold or traded), and the market capitalization mi (the market value as of a publicly traded company’s outstanding shares) at time i, have also been normalized to be in the interval [0,1].
3.4 Perturbation measures
We considered perturbations coming from different sources for the stock market and for the cryptocurrencies market, according to the data we analyzed, all of them internal in accordance with the discussion in Section 2.2.2.
Essentially, we associate perturbations with fluctuations or volatility. So, we computed this measure by assessing how much the assets changed their parameters (price and volume in the case of stocks, and also market capitalization in the case of cryptocurrencies) in two consecutive observations.
In the case of stocks, as our data did not include values of market capitalization, we included other indicators, such as the volatility index, with symbol VIX, and the mean of the volatility of Nasdaq, Dow Jones and S&P 500 indexes.
More concretely, let p(x)i, v(x)i and m(x)i be the price, volume, and market capitalization of an asset x at time i, respectively. Then, in the case of stocks, we defined the different perturbations as:
(7)
(8)
(9)
- Pvix(i) is the value of the volatility index VIX at time i.
- P3m(i) is the average of absolute changes of the Nasdaq, Dow Jones and S&P 500 indexes between time i-1 and time i. That is, if N(i), D(i), &(i), are the normalized values corresponding to such indices at time i, then:
Note that all three indices were normalized using Eq (5), so there is no difference in their orders of magnitude.
Finally, the perturbation measure for stocks is defined as the mean value of these four perturbation sources:
(11)
In the case of cryptocurrencies we considered:
(12)
(13)
(14)
(15)
The values of Ppr(x,i) and Pnp(x,i) are different in the sense that, for Pnp(x,i), the prices were normalized before taking the mean over all the assets and, for Ppr(x,i), afterwards. So the perturbation contributed by an asset in the former measure is equal to another asset of different price if they changed the same percentage of their price. While in the latter, the perturbation of the system is governed by expensive cryptocurrencies.
Correspondingly, the definition of the perturbation measure is defined as the mean value of these four perturbation sources.
3.5 Computation of antifragility
We computed antifragility values for every asset in each system considering different time-scales: daily, weakly, monthly and yearly–the latter only plotted in the case of stocks. As we will see in the following sections, the behavior of our antifragility measure does not change drastically between time-scales. This observation is consistent with previous findings on the distribution of stock and cryptocurrencies returns [38, 50–52]. We calculate the antifragility using Eqs (1) and (2), considering the perturbation given by Eq (11) (for stocks) and Eq (16) (for cryptocurrencies).
3.6 Performance measures
We compared quantitatively the antifragility values with different “performance measures” listed next:
- age: age of the asset in days.
- pct_dlt-pr: maximum price minus minimum price divided by the mean price.
- pct_dlt-mk: same as before but using market capitalization instead of price.
- pct_dlt-vl: same as before but using volume.
- pct_pr-f-i: final price minus initial price divided by the mean price.
- pct-mk-f-i: same as before but using market capitalization instead of price.
- pct-vl-f-i: same as before but using volume.
- pr_mea: mean price.
- pr_std: price standard deviation.
- mk_mea: mean market capitalization.
- vl_mea: mean volume.
We recall that the variables which involve the market capitalization were only used in the case of cryptocurrencies, as we do not have this data for stocks. Instead, we used lists of the best stocks of the year (from 2010 to 2017) according to different stock market analysts, such as Forbes, Yahoo Finance, Stock Market Watch, among others. We will often refer to the stocks belonging to such lists as the ‘top-performers’, and to the variables listed above as the ‘good-performance’ measures.
4. Results and discussion
As an example, in Fig 1 we show, for two different cryptocurrencies (ethereum and bitcoin), a time series for perturbation, satisfaction (return) and antifragility. Notice that the time scales are different. We used normalized values as explained above.
Notice that the time intervals are different. We calculated the satisfaction (price return) using Eq (1) for the discrete time i = 1,2,…,N; and the antifragility using Eq (2). For the perturbation in Eq (2) we use the average quantity for cryptocurrencies given by Eq (16). We plot this perturbation as a third curve.
4.1 Antifragility in the stock and cryptocurrency markets
In what follows, we will present a set of figures where we analyze the data of assets for the stock and cryptocurrency markets. We will show the values of antifragility versus the different performance measures listed above in Section 3.6.
As a first example, we plot in Fig 2 the antifragility A, that we indicated as “aft”, against one of the performance measures. In this case, we used the mean price (pr_mean) of several cryptocurrencies. The mean price is evaluated considering the full range of the discrete time in the data set. We consider weekly prices here. We are plotting a green dot for each type of cryptocurrency and we have around 1800 different cryptocurrencies in the data set. We only indicate in the figure the name of some of the cryptocurrencies for clarity. It is worth remembering here the meaning of antifragility A: we consider that an asset is robust if A is close to 0, fragile if it is negative, and antifragile if it is positive. Notice that the dots are centered around zero with a considerable dispersion.
We compare the antifragility values (in the x-axis) of every cryptocurrency in a weekly scale (according to the green color), with its mean price, indicated by pr_mea, (in the y-axis). Note that in this case the symlog scale is just a log scale because there are no negative values in the y-axis.
We now show a series of similar figures of antifragility vs different performance measures. We will consider different time scales for the antifragility, distinguished with a corresponding color: daily in cyan, weekly in green, monthly in red and semiannual (only included in the case of stocks) in yellow. In Figs 3–6 and 8–11, the horizontal axis represents antifragility and the value of A is centered around zero. In these figures, each dot represents a different asset in the stock market, and we indicate inside each figure just a few of their names for clarity. The number of stocks analyzed is about 7000.
Each subplot corresponds to a timescale (daily in cyan, weekly in green, monthly in red, semiannual in yellow). The plots compare the antifragility A measure (in the x-axis) to the corresponding ‘good-performance’. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0. See the text for details of this figure.
Each subplot corresponds to a timescale (daily in cyan, weekly in green, monthly in red, semiannual in yellow). The plots compare the antifragility A measure (in the x-axis) to the pct_dlt-vol of each asset. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0.
Each subplot corresponds to a timescale (daily in cyan, weekly in green, monthly in red, semiannual in yellow). The plots compare the antifragility A measure (in the x-axis) to the pr_mea of each asset. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0.
Each subplot corresponds to a timescale (daily in cyan, weekly in green, monthly in red, semiannual in yellow). The plots compare the antifragility A measure (in the x-axis) to vl_mea of each asset. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0.
We will start with Figs 3–6. Each of these figures contains four plots representing different time scales. In Fig 3, we are plotting the age (measured in days) of the asset in question. In Fig 4, we show the maximum volume minus the minimum volume divided by the mean volume (indicated by “pct_dlt-vol”). In Fig 5, we plot the mean price, and in Fig 6, we show the mean volume.
About this figure we can say the following:
- The qualitative behavior of the antifragility A is the same for different times scales: daily (cyan), weekly (green), monthly (red) and semiannual (yellow).
- There are no linear correlations between any of the ‘good-performance’ measures and the antifragility measures.
- The higher values of the ‘good-performance’ measures concentrate around the robust (A = 0) axis. We used different scales in order to better illustrate this feature.
- The behavior pointed out in (3) is clearer and more consistent for different time scales among the ‘good-performance’ measures in all of these figures.
We now plot, in Fig 7, the probability distributions of the antifragility for different time scales, indicated by different colors as shown in the inset. Notice that these distributions are close to normal distributions.
The dashed line represents the mean antifragility of the set of ‘top-performers’. We observe that the mean A of ‘top-performers’ is only greater than the mean A in the daily timescale.
Now we turn our attention to the case of the cryptocurrency market.
In Figs 8–11, we carried on essentially the same analysis as in the case of stocks. Among the following plots, colors represent a time scale: daily in cyan, weekly in green, monthly in red.
Each subplot corresponds to a timescale. The plots compare the corresponding A measure (x-axis) to the ‘good-performance’ measure age. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0.
Each subplot corresponds to a timescale. The plots compare the corresponding A measure (x-axis) to the ‘good-performance’ measure pct_dlt-vl. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0.
Each subplot corresponds to a timescale. The plots compare the corresponding A measure (x-axis) to the ‘good-performance’ measure pr_mea. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0.
Each subplot corresponds to a timescale. The plots compare the corresponding A measure (x-axis) to the ‘good-performance’ measure pct_dlt-pr. We observe how higher values of the ‘good-performance’ measure concentrate in the robust axis A = 0.
In Fig 8, we are plotting the age that corresponds to the age in days of the asset in question. In Fig 9, we show the maximum volume minus the minimum volume divided by the mean volume (indicated by “pct_dlt-vol”). In Fig 10, we plot the mean price. Finally, in Fig 11, we plot the same as in Fig 9 but with price instead of volume (indicated by “pct_dlt-pr”).
We observe a similar behavior as in the case of stocks. The previous observations (1–4) also hold in this case. And it can be observed a greater concentration of assets on the ‘fragile’ side.
Finally, the probability distributions of the antifragility measures, shown in Fig 12, are close to skew-normal distributions, with a left asymmetry towards lower values of antifragility. The probability distributions of the antifragility values of the ‘top-performers’ seem to slightly improve from the rest. See Fig 12.
The dashed line represents the mean antifragility of the set of ‘top-performers’. We observe that the mean A of ‘top-performers’ is greater than the mean A in every timescale.
5. Discussion and conclusions
Antifragility occurs when systems benefit from perturbations. This is opposite to fragility, where noise or changes damage a system. The absence of fragility, when perturbations do not affect a system, is robustness. In the introduction of this paper, we briefly mentioned some systems that benefit in the presence of noise. We mentioned some phenomena that have been thoroughly studied recently: stochastic resonance, noise-induced transport and optimization by simulated annealing, among others. All these phenomena share the property of gaining in the presence of a stochastic signal or perturbation, thermal noise or being exposed to other fluctuations.
The concept of antifragility can be related to another concept: persistence. Imagine that a random variable fluctuates starting from an initial condition that defines a threshold of the system. The persistence time is defined as the time that the variable takes before crossing this threshold. This concept is closely related to the first-passage time and with the survival probability. It is clear that there might be a connection between these two concepts: antifragility and persistence. If we think of a system with a threshold, then an important quantity to study is precisely the persistence probability or the persistence time as a random variable. For instance, this threshold can be related to the breaking point of a fragile material, the limited stress in a seismic system before a big earthquake occurs, or the critical value of a financial system before a crash takes place. One can think of many other examples where the concept of persistence can be useful: earth and atmospheric sciences, complex networks, economics and market analysis, non-equilibrium systems, statistical physics, optimization and planning, energy resources and health sciences (see for instance the recent review [53])). For applications in physics see [54] and the review [55]; for economics and finance see [56, 57]. The main idea for connecting persistence with antifragility is to equate both concepts, that is, the antifragility is proportional to the persistence probability. Another possibility can be that antifragility is proportional to the persistence time in a time series. The longer the time below the breaking threshold, the longer the antifragility of the system and the longer the persistence time. We think that this connection opens interesting and promising avenues of research for future work.
As mentioned in the introduction, other important examples of complex dynamical systems that might have antifragility, are economic and financial systems. Clearly, both are complex systems that are open, nonlinear, and have many agents that operate collectively, giving rise to emergent properties like the ones we observe in financial markets and other economic indicators.
An important question that we may ask is: can we identify the phenomenon of antifragility in economics and finance? As far as we know, this question has not been answered in a definite way and, worse than that, the question usually is not even asked or posed in these terms. Therefore, the main goal of this paper is: firstly, to pose this question explicitly and, secondly, to try to answer it by proposing a novel measure of antifragility for stocks and cryptocurrencies markets.
Our proposal is to define explicitly the antifragility in mathematical terms as a multiplicative stochastic process. Our definition is simple: antifragility is the product of returns and perturbations. Based on this definition, we analyze the returns to be used: price returns of stocks and cryptocurrencies in the market. As perturbations, we used different options that come from the market itself, like fluctuations of price returns, volume, market capitalization and global average indexes. One can think of these perturbations as endogenous shocks. We have not studied here the case of exogenous shocks or aggregate shocks, but it could be interesting to analyze these cases in a further analysis for the future. Finally, with these returns and perturbations, we calculated the antifragility measure. Both the returns and the perturbations were obtained from data sets that are publicly available, as indicated in the text.
Once we obtain the antifragility, we explore its dependence against different performance measures: average quantities related with price, volume, and market capitalization.
We would also like to stress the fact that we carried on numerous experiments with different computations of antifragility, particularly about the choices of perturbation measures. In early stages of our research we analyzed a different measure of antifragility associated with each type of perturbation individually (see Eqs 7–9, and 12–15). We also carried on time-sectional analysis by comparing the data among different years and months. In the big picture, all of these experiments exhibited essentially the same behavior, both among them and to the one described in the current version of the manuscript as well. For this reason we focused only on the latter. It is worth mentioning that, by defining antifragility in the way we did, there are first-order effects that can be seen in a set of assets whose antifragility and satisfaction (price returns) time series are highly correlated (either positively or negatively). This happens for different reasons, either they are very new assets and so there are few observations, or they had almost no change in some of their parameters such as a constant price or volume.
As a matter of conclusion of this study, we can point out the following:
The results obtained with our definition of antifragility indicate that both the stock and the cryptocurrencies markets do not show an antifragile character in a clear way and seem to be favored by robustness instead. There were no correlations between our antifragility measures and the different performance measures that we analyze, while the best performance assets seem to fall within the category of robustness. In virtually every analysis we carried out, the highest and lowest antifragility values were obtained for the newest assets and for those with the lowest transaction volume. For example, the average age of the top ~300 antifragile stocks is 302 days, the oldest of them is less than 3 years old, and over 80% are younger than a year. Moreover their mean volume of transactions is around 5 times smaller than the mean volume of transactions overall. It can be observed that, in both scenarios, the assets tend to evolve towards robustness. In Figs 13–15 we show this feature by plotting the cumulative mean of the antifragility measure over time. In Fig 13, it is shown this evolution for a selected set of ‘top-performing’ stocks in a monthly scale for about twelve years; Fig 14 shows the same time scale and span but with a random set of assets; and Fig 15 shows a daily evolution of a selected set of cryptocurrencies for one year.
The x-axes shows the date and the y-axis the cumulative antifragility measure. The figure shows the cumulative monthly antifragility of selected ‘top-performing’ stocks.
The x-axes shows the date and the y-axis the cumulative antifragility measure. The figure shows the cumulative monthly antifragility of 25 random stocks.
The x-axes shows the date and the y-axis the cumulative antifragility measure. The figure shows a daily evolution of a selected set of cryptocurrencies for one year.
Finally, it is important to stress that these results depend obviously on the definition of antifragility that we used. It would be interesting to address other possible definitions that capture the essence of the phenomenon. Meanwhile, we sincerely hope that this work stimulates the interest of researchers to study antifragility as an important feature in financial systems and in complex dynamical systems in general.
Acknowledgments
We are grateful to Ewan Colman, Hyobin Kim and Dante Pérez for interesting discussions and suggestions; and to the C3 community, specially to José Luis Gordillo for providing technical support with the Center’s Cluster usage.
References
- 1.
N. N. Taleb, “Antifragile: Things That Gain from Disorder”, Random House, New York, NY, USA, 2012.
- 2. Aven T. (2015). The concept of antifragility and its implications for the practice of risk analysis. Risk analysis, 35(3), 476–483. pmid:25263809
- 3. Cavanagh C. J. (2017). Resilience, class, and the antifragility of capital. Resilience, 5(2), 110–128.
- 4. White L. H. (2013). Antifragile banking and monetary systems. Cato J., 33, 471.
- 5. de Bruijn H., Größler A., & Videira N. (2020). Antifragility as a design criterion for modeling dynamic systems. Systems Research and Behavioral Science, 37(1), 23–37.
- 6. Blečić I., & Cecchini A. (2017). On the antifragility of cities and of their buildings. City, Territory and Architecture, 4(1), 3.
- 7. Johnson J., & Gheorghe A. V. (2013). Antifragility analysis and measurement framework for systems of systems. International Journal of Disaster Risk Science, 4(4), 159–168.
- 8. Taleb N. N., & Douady R. (2013). Mathematical definition, mapping, and detection of (anti) fragility. Quantitative Finance, 13(11), 1677–1689.
- 9. Taleb N. N., & Douady R. (2011). A map and simple heuristic to detect fragility, antifragility, and model error. NYU-Poly working paper, SSRN.
- 10. Taleb M. N. N., Canetti M. E., Kinda M. T., Loukoianova M. E., & Schmieder M. C. (2012). A new heuristic measure of fragility and tail risks: application to stress testing. International Monetary Fund.
- 11. Pineda Omar K., Kim Hyobin, and Gershenson Carlos (2019). “A Novel Antifragility Measure Based on Satisfaction and Its Application to Random and Biological Boolean Networks,” Complexity, vol. 2019, Article ID 3728621.
- 12. Kim H., Pineda O. K., & Gershenson C. (2019). “A Multilayer Structure Facilitates the Production of Antifragile Systems in Boolean Network Models”. Complexity, vol. 2019, Article ID 2783217.
- 13. Kim H., Muñoz S., Osuna P., & Gershenson C. (2020). “Antifragility Predicts the Robustness and Evolvability of Biological Networks through Multi-class Classification with a Convolutional Neural Network”. arXiv preprint 2002.01571, 2020.
- 14. Equihua M., Espinosa Aldama M., Gershenson C., López-Corona O., Munguía M., Pérez-Maqueo O., and Ramírez-Carrillo E. (2020). Ecosystem antifragility: beyond integrity and resilience. PeerJ, 8:e8533. pmid:32095358
- 15. Gammaitoni L., Hänggi P., Jung P., and Marchesoni F. Stochastic resonance. Reviews of Modern Physics 70 (1) (1998) 223.
- 16. Peter Hänggi and Fabio Marchesoni. Artificial Brownian motors: Controlling transport at the nanoscale. Reviews of Modern Physics 81 (1) (2009) 387.
- 17. José L. Mateos and Alatriste F. R. Brownian motors and stochastic resonance. Chaos 21 (2011) 047503. pmid:22225377
- 18. Kirkpatrick S., Gelatt C. D. Jr., and Vecchi M. P. Optimization by Simulated Annealing. Science 220 (4598) (1983) 671. pmid:17813860
- 19.
Dowsland K. A. and Thompson J. Simulated Annealing. In, Rozenberg G., Back T.and Kok J. N.editors. Handbook of Natural Computing. Springer-Verlag, Berlin (2012).
- 20.
Pikovsky Arkady, Rosenblum Michael and Kurths Jürgen. Synchronization: a universal concept in nonlinear science. Cambridge University Press (2003).
- 21.
Von Foerster H. (1960). "On self-organizing systems and their environments", pp. 31–50 in Self-organizing systems. Yovits M.C. and Cameron S. (eds.), Pergamon Press, London
- 22. Atlan H. (1974). On a formal definition of organization. Journal of Theoretical Biology, 45(2):295–304. pmid:4844620
- 23. Black F. Noise. Journal of Finance 41(3), 529–543 (1986).
- 24. Mahata A., Rai A., Nurujjaman M., and Prakash O., “Modeling and analysis of the effect of COVID-19 on the stock price: V and L-shape recovery,” Physica A 574, 126008 (2021). pmid:36568062
- 25. Ramezani J., Camarinha-Matos L. M., “Approaches for resilience and antifragility in collaborative business ecosystems”. Technological Forecasting and Social Change 151, 119846(2020), ISSN 0040-1625.
- 26. Nikookar E., Varsei M., Wieland A., “Gaining from disorder: Making the case for antifragility in purchasing and supply chain management”, Journal of Purchasing and Supply Management 27–3, 100699(2021), ISSN 1478-4092, https://doi.org/10.1016/j.pursup.2021.100699.
- 27.
Kahneman, D., Sibony, O., and Sunstein, C. R. Noise: A flaw in human judgment. Little, Brown Spark, New York, USA (2021).
- 28.
Sornette D. Why Stock Markets Crash: Critical events in complex financial systems. Princeton University Press, Princeton, USA (2003).
- 29. Cutler D., Poterba J. M., and Summers L. H. What moves stock prices? The Journal of Portfolio Management, 15(3), 4–12 (1989).
- 30. Gabaix X. The Granular Origins of Aggregate Fluctuations. Econometrica 79(3), 733–772 (2011).
- 31. Gabaix X. PowerLaws in Economics: An Introduction. Journal of Economic Perspectives 30(1), 185–206 (2016).
- 32.
Mantegna R. N. and Stanley H. E. An Introduction to Econophysics: Correlations and Complexity in Finance. Cambridge University Press, Cambridge, UK. (2000).
- 33. Kutner R., Ausloos M., Grech D., Di Matteo T., Schinckus C., and Stanley H. E. Econophysics and sociophysics: Their milestones & challenges. Physica A 516 (2019) 240–253.
- 34. Buonocore R. J., Musmeci N., Aste T., and Di Matteo T. Two different flavours of complexity in financial data. Eur. Phys. J. Special Topics 225, 3105–3113 (2016).
- 35. Pan R. K. and Sinha S. Collective behavior of stock price movements in an emerging market. Physical Review E 76(4), 046116 (2007).
- 36. Yadav G. S., Guha A. and Chakrabarti A. S. Measuring complexity in financial data. Frontiers in Physics, 8 (339) (2020).
- 37. Alves L. G. A., Sigaki H. Y. D., Perc M., and Ribeiro H. V. Collective dynamics of stock market efficiency. Scientific Reports 10, 21992 (2020). pmid:33319788
- 38. Pogudin A., Chakrabati A. S. and Di Matteo T. Universalities in the dynamics of cryptocurrencies: stability, scaling and size. Journal of Network Theory in Finance, 5(4), 63–95 (2019).
- 39. Alessandretti L., ElBahrawy A., Aiello L. M., and Baronchelli A. Anticipating Cryptocurrency Prices Using Machine Learning. Complexity 2018, 8983590 (16 pp.) (2018).
- 40. Wu K., Wheatley S., and Sornette D. Classification of cryptocurrency coins and tokens by the dynamics of their market capitalizations. Royal Society Open Science 5, 180381 (2018). pmid:30839686
- 41. James N., Menzies M, Chan J., “Changes to the extreme and erratic behaviour of cryptocurrencies during COVID-19”, Physica A 565, 125581 (2021). pmid:33250564
- 42. Nie Chun-Xiao, “Analysis of critical events in the correlation dynamics of cryptocurrency market”, Physica A: Statistical Mechanics and its Applications, 586, 126462(2022), ISSN 0378-4371, https://doi.org/10.1016/j.physa.2021.126462.
- 43. Luo M., Kontosakos V. E., Pantelous A. A., Zhou J., “Cryptocurrencies: Dust in the wind?”, Physica A: Statistical Mechanics and its Applications, 525(2019), ISSN 0378-4371, https://doi.org/10.1016/j.physa.2019.03.123.
- 44. Bardoscia M., Barucca P., Battiston S., Caccioli F., Cimini G., Garlaschelli D., et al. The Physics of Financial Networks. Nature Reviews Physics, 3, 490–507 (2021).
- 45. Arthur W. B. Foundations of Complexity Economics. Nature Reviews Physics, 3, 136–145 (2021). pmid:33728407
- 46. Hidalgo C. A. Economic Complexity Theory and Applications. Nature Reviews Physics, 3, 92–113 (2021).
- 47.
Hidalgo C. A. Why Information Grows: The Evolution of Order from Atoms to Economies. Basic Books, New York, USA (2017).
- 48. Marjanovic, Boris (2018), Historical daily prices and volumes of all U.S. stocks and ETFs, Huge Stock Market Dataset, available at: https://www.kaggle.com/datasets/borismarjanovic/price-volume-data-for-all-us-stocks-etfs, (accessed March 2022).
- 49. Vent Jese (2018), Daily crypto markets open, close, low, high data for every token ever, Every Cryptocurrency Daily Market Price, available at: https://www.kaggle.com/datasets/jessevent/all-crypto-currencies, (accessed March 2022).
- 50. Mandelbrot B. (1963). The Variation of Certain Speculative Prices. The Journal of Business, 36(4), 394–419. Retrieved April 30, 2021, from http://www.jstor.org/stable/2350970.
- 51. Gopikrishnan P., Plerou V., Amaral L. A. N., Meyer M., & Stanley H. E. (1999). Scaling of the distribution of fluctuations of financial market indices. Physical Review E, 60(5), 5305. pmid:11970400
- 52. Gabaix X., Gopikrishnan P., Plerou V., & Stanley H. E. (2003). A theory of power-law distributions in financial market fluctuations. Nature, 423(6937), 267–270. pmid:12748636
- 53. Salcedo-Sanz S., Casillas-Pérez D., Del Ser J., Casanova-Mateo C., Cuadra L., Piles M., et al. “Persistence in complex systems”. Physics Reports 957 (2022) 1–73.
- 54. Majumdar S. N., Bray A. J., Cornell S. J., and Sire C. “Global Persistence Exponent for Nonequilibrium Critical Dynamics”. Physical Review Letters 77 (1996) 3704.
- 55. Bray Alan J., Majumdar Satya N., and Schehr Grégory. “Persistence and first-passage properties in nonequilibrium systems”. Advances in Physics 62 (2013) 225–361.
- 56. Constantin M. and Das Sarma S. “Volatility, persistence, and survival in financial markets”. Physical Review E 72 (2005) 051106.
- 57. Roberto da Silva Marcelo Zembrzuski, Correa Fabio C., and Lamb Luis C. “Stock market and criticality in the current economic crisis”. Physica A 389 (2010) 5460–5467.