Advanced Studies of Financial Technologies and Cryptocurrency Markets

Cryptocurrencies—digital assets—are discussed from the viewpoint of the medium of exchange and the store of value with a focus on Bitcoin. The issue of trust towards the unit of accounts is viewed in the perspective of historical events including fiat currency reforms in the past all over the world. It is argued that a major test of the present global financial system and its emerging new technology alternatives has not occurred yet. The notion of intrinsic value of currency is strongly attached to the stability of the social and economic system, which landscape will be probably drastically altered in the next few decades due to the rise of artificial intelligence. The chapters in this book which are focusing on the current aspects of financial innovations also point out the possible directions and dilemmas the future may bring us.

In a financial network where mark-to-market accounting rules apply, the sale of assets enforced by behavioral constraints such as minimum capital requirements can induce an amplification effect of additional asset sales that further depresses the market price. This paper explores these contagious processes through simulation exercises under some different sets of network structures. We introduce a complete graph, clusters and core periphery while varying the composition of banks’ portfolios and observing their effects on outbreaks and the spread of a financial contagion. This paper also investigates ex ante conditions that could prevent a contagion and examines some ex post measures that could restrain the propagation of a contagion. Securing a certain level of liquidity in a financial system that includes large-scale banks can be an effective ex ante regulatory measure. Additionally, certain ex post operations, such as a price-supporting purchase of risky assets and/or a capital injection into a bank, could be effective countermeasures to prevent the contagion from spreading under some limited conditions.

Recent research has seen increasing use of risk/diversification based approach to portfolio optimization. Under the risk-based approach, returns are ignored, and a diversification or risk measure is optimized in portfolio construction. This approach has been criticized for lacking a clearly defined objective like the trade-off between returns and risk in the Markowitz’s Mean-variance set up. Optimizing risk/diversification alone is a single objective optimization approach to portfolio construction. This is in contrast to the usual bi-objective optimization that yields the portfolio that optimizes the trade-off between return and risk. In this paper, we note that portfolios that optimize the trade-off between returns and diversification measures exist (i.e. portfolios that replace variance with other risk measures). In theory, these mean-diversification optimal portfolios should dominate risk-based portfolio on a risk-adjusted basis. Using genetic algorithm, mean-diversification efficient frontiers are drawn for various diversification measures and portfolios that optimize the trade-off between returns and the diversification measures are identified. We argue that these portfolios are better candidates to be compared with the portfolio that is constructed to be mean-variance optimal since they sensitive to returns. Our results suggest that mean-diversification optimal portfolios are useful alternatives in terms of risk-reward trade-off based on their in-sample and out-of-sample performance.

Forecasting financial time series has been traditional and important theme for market analysis and investment strategy. However, it is not easy to capture the statistical characteristics of the data due to high noise level and volatile features. On the other hand, technological innovation by artificial intelligence is progressing rapidly in various fields. Especially, long short-term memory (LSTM) has been widely used in natural language processing and speech recognition. In this paper, we study prediction performance of LSTM by comparing it with other machine learning models such as logistics regression and support vector machine. The characteristics of these models were first investigated by applying them to predict different types of simulated time series data. We then conducted an empirical study to predict stock returns in TOPIX Core 30 with application to portfolio selection problem. Overall, LSTM showed favorable performance than other methods, which is consistent with Fischer and Krauss (Eur J Oper Res 270(2):654–669, 2018) for S&P500 data.

We study contagious defaults of banks by applying a voting model. The network of banks are created by the relation, lending and borrowing among banks. We introduce the response function from Merton model. Using this response function we calculate the probability of default (PD) which includes not only changes of asset values but also the effects of connected banks’ defaults using the mean field approximation. If we approximate the normal distribution which Merton model uses by \(\tanh \) function, we can obtain the kinetic Ising model which represents phase transition. The asset volatility plays the role of temperature. In the low temperature limit, the model becomes the threshold model. We calculate PD which shows the effect of the situations around the bank as the additional PD using the self consistent equation.

After several years of the proposal and implementation of Bitcoin by Satoshi Nakamoto, people in the world were enthusiastic about crypto-assets. However, the market prices of crypto-assets are too unstable to use as a payment method. After many cyber-attack incidents, the confidence in the security of crypto-asset exchanges has also been compromised. Satoshi proposed Bitcoin to realize anonymous payment to protect individual privacy. Actual crypto-assets have changed from the original concept. The main reason for this deviation was the reality that ordinary investors cannot manage their secret keys securely. In this chapter, the reasons for this deviation are investigated.

How money flows among users of Bitcoin is an interesting question in order to understand the dynamics on the complex network of Bitcoin transactions, and also how the transactions are related to the price in the exchange market indirectly. We employ the data of blockchain in the Bitcoin from 2013 to 2018 (compiled by a Hungary research group), utilize a simple algorithm to partially identify anonymous users from addresses, and construct snapshots of temporarily changing network with the users as nodes and the transactions as directed links. In order to understand how users are located in the entire flow, in particular upstream and downstream, we use the so-called Hodge decomposition (or Helmholtz-Hodge-Kodaira decomposition). In addition, we examine the so-called “bow-tie” structure of the binary network disregarding flow to find how the users in the upstream/downstream peripheries (the so-called IN/OUT) are located away from the core of strongly connected component (SCC). We compare the Hodge potential of each node with such a location in the bow-tie structure, and also with the net demand/supply of each node measured from the money flow, to find a significant correlation among the potential, the topological position, and the net demand/supply. We also decompose the flow of each link into potential flow and circular flow to find that circulation of the flow is quite significant. We shall discuss about possible implication of these findings.

There are several previous empirical analyses for Bitcoin pricing; however, only a few pieces of research can be found in terms of the relationships among major crypto assets, such as Ethereum, and Ripple. Here, we apply a method proposed by Nan and Kaizoji (Int Rev Fin Anal 64:273–281, 2019), which calculates an indirect exchange rate to consider the possibility of a cointegrating relationship between a crypto-asset exchange rate and a direct FX spot rate. We investigate market efficiency in crypto-asset exchange rates through the application of several kinds of unit root tests and the Johansen procedure. The results suggest that the weak form of market efficiency does not seem to hold for all pairs; however, one of the prerequisites for semi-strong form of market efficiency holds for several pairs. Additionally, we focus on the dynamic relationships by applying the impulse response function for a four-variable VECM. Remarkably, the Bitcoin exchange rate can slightly affect the \( EUR/USD \) spot rate.

This study proposes utilizing FX futures to hedge the risk of currency exchanges based on the bitcoin exchange rate. The time-dependent optimal hedge ratio for the resulting portfolio can be calculated from the conditional covariance matrix of the two returns. To model the conditional joint density, a VECM plus DCC-GARCH model is suggested due to the existence of co-integration between the bitcoin exchange rate and FX futures. Comparisons suggest that this framework is superior to the commonly used naïve and conventional hedging strategies in several important aspects.

The Ether cryptocurrency, based on the blockchain of the Ethereum project for smart contracts, has long had the 2nd market capitalization, next to the Bitcoin. Despite its importance and the innovative features of the entire Ethereum ledger ecosystem, Ether has attracted far less attention than Bitcoin in terms of the time series analysis. This work provides an analysis of the R/S Hurst Exponent for the Ether time series in order to test to what extent the price dynamics may be predictable by deterministic methods including machine learning. Daily log returns, volatility time series, and transaction count sequences are analyzed. Support Vector Machine algorithm is used for testing the marginal predictability level. Ether-mediated triangular arbitrage between six major fiat currencies is also studied—we provide the distributions of the logarithmic rate of arbitrage transaction return for the 15 currency pair combinations. We also study the cointegration process of Ether-exchange rates with the foreign exchange rates that are the cause and driving force of the adjustment process towards dynamic market equilibrium eliminating arbitrage windows. The efficiency of the Ether market is found to increase with time.

The proportion of informed traders in financial markets is seen as a measure for the degree of information asymmetry and has been used to explain the existence of Bid-Ask spread. We identify a proxy—a spread-to-range ratio—for the unobserved proportion of informed traders in a market from the classic Glosten-Milgrom (1985) model. It can be shown that this ratio is the minimum of the proportion of informed traders, and the respective dynamics of spread and range motivate the conditional modelling of the ratio. Empirical results are given for the BTC-USD data over an 1186-day period, which indicate that the estimated proportion of informed traders can be as high as 6% in the cryptocurrency market.

Our study is devoted to the problems of the short-term forecasting cryptocurrency time series using machine learning (ML) approach. Focus on studying of the financial time series allows to analyze the methodological principles, including the advantages and disadvantages of using ML algorithms. The 90-day time horizon of the dynamics of the three most capitalized cryptocurrencies (Bitcoin, Ethereum, Ripple) was estimated using the Binary Autoregressive Tree model (BART), Neural Networks (multilayer perceptron, MLP) and an ensemble of Classification and Regression Trees models—Random Forest (RF). The advantange of the developed models is that their application does not impose rigid restrictions on the statistical properties of the studied cryptocurrencies time series, with only the past values of the target variable being used as predictors. Comparative analysis of the predictive ability of the constructed models showed that all the models adequately describe the dynamics of the cryptocurrencies with the mean absolute persentage error (MAPE) for the BART and MLP models averaging 3.5%, and for RF models within 5%. Since for trading perspective it is of interest to predict the direction of a change in price or trend, rather than its numerical value, the practical application of BART model was also demonstrated in the forecasting of the direction of change in price for a 90-day period. To this end, a model of binary classification was used in the methodology for assessing the degree of attractiveness of cryptocurrencies as an innovative financial instrument. Conducted computer simulations have confirmed the feasibility of using the machine learning methods and models for the short-term forecasting of financial time series. Constructed models and their ensembles can be the basis for the algorithms for automated trading systems for Internet trading.

This study examines the relationship between the return on Bitcoin and the returns on its forks (Litecoin, Bitcoin Cash, Bitcoin Gold, Bitcoin Diamond, and Bitcoin Private). I obtain volatility series and time-varying correlation coefficients (Bitcoin with each of its forks) based on both univariate and multivariate GARCH models (EWMA, DCC, and BEKK). In terms of volatility, the gains of using a multivariate volatility approach are not substantial. However, the three multivariate volatility models offer a better estimation of the time-varying correlation. This study provides evidence that the volatility of Bitcoin forks and the volatility of Bitcoin are dynamically related, and there is a transmission of volatility risk from Bitcoin forks to Bitcoin. The results suggest that Bitcoin and its forks behave as crypto-currencies during bad times and as assets during good times. Also, for most of the sample period, Bitcoin forks do not offer a hedge against Bitcoin risk.