MDVsoft
commented
2 years ago Bitcoin: Learning and Predictability via Technical Analysis∗ Andreas Bernanek University of Vienna Hong Liu Washington University in St. Louis Jack Strauss University of Denver Guofu Zhou† Washington University in St. Louis Yingzi Zhu Tsinghua University December 10, 2018 Abstract We document that 1- to 20-week moving averages (MAs) of daily prices predict Bitcoin returns in- and out-of-sample. Trading strategies based on MAs generate substantial alpha, utility and Sharpe ratios gains, and significantly reduce the severity of drawdowns relative to a buy-and-hold position in Bitcoin. We explain these facts with a novel equilibrium model that demonstrates, with uncertainty about growth in fundamentals, rational learning by investors with different priors yields predictability of returns by MAs. We further validate our model by showing the MA strategies are profitable for tech stocks during the dotcom era when fundamentals were hard to interpret. JEL classification: G11, G12, G14 Keywords: Bitcoin, cryptocurrency, technical analysis
- Introduction Relative to other assets, the historical returns on Bitcoin are astounding. One dollar invested in Bitcoin on July 18, 2010 grew to $70,970 by June 30, 2018, after hitting a peak value of $214,922. The same investment in the S&P500 stock index, which experienced relatively strong performance by historical standards, grew to only $2.55 over the same period. Moreover, Bitcoin is only the first digital coin in the rapidly growing cryptocurrency market whose capitalization measures in the hundreds of billions of dollars. Bitcoin’s dramatic price growth accompanies similarly dramatic volatility. For example, Bitcoin experienced three large drawdowns averaging 30% in 2017 alone. Few, if any, measurable fundamentals explain Bitcoin’s explosive price growth and high volatility. In this paper, we address two natural asset pricing questions: what explains the dynamics of Bitcoin prices and are Bitcoin returns predictable? Presumably, the value of Bitcoin depends on some form of convenience yield, e.g., the benefits it provides as a medium of exchange and store of value. However, unlike more established governmentbacked currencies, Bitcoin’s current and expected future convenience yield is a highly uncertain and randomly evolving quantity investors must learn about. Furthermore, unlike other common financial assets such as stocks, bonds, and currencies, Bitcoin lacks universally accepted valuerelevant fundamentals such as dividends, interest payments, and accounting statements. Given this lack of fundamentals, one naturally suspects that Bitcoin investors rely heavily on technical indicators such as the path of prices and volume to infer value-relevant information. In this paper, we provide the first study of technical analysis applied to Bitcoin. We also construct a novel rational continuous-time equilibrium model that justifies the use of common technical analysis strategies, not only in trading Bitcoin, but also in trading financial assets that lack easily interpreted fundamentals.1 In the model, rational investors with different priors learn about the latent growth rate of Bitcoin’s convenience yield. In contrast to other financial assets, investors in our model can only learn from past prices and convenience yields; they do not have access to other value-relevant signals like fundamentals. We show that, in this setting, trading rules based on moving averages (MAs) are optimal, MAs predict returns, and price drift exists (i.e., the expected return increases after a rise in Bitcoin’s price). Unlike prior models with technical traders, our model does not assume that a subset of investors use exogenously given technical trading rules (e.g., Han et al., 2016). Relative to the theoretical literature on technical analysis, our model is the first equilibrium model with endogenous technical trading. Our model also differs from prior rational models that generate time-series momentum-style price drift. For example, in the model of Banerjee et al. (2009), the existence of price drift relies on differences in higher-order beliefs, and in the model of Cochrane et al. (2008), this drift requires existence of multiple risky assets.2 Consistent with our model, we find that daily Bitcoin returns are predictable in-sample by ratios of current (log) price to its 1- to 20-week moving averages (MAs). The model also predicts that this predictability becomes stronger when uncertainty decreases as investors learn about the dynamics of the latent growth of the convenience yield. Consistent with this prediction, we find both a negative interaction between the price-to-MA and return variance in return-forecasting regressions, as well as a downward trend in variance over time. Since in-sample predictive regressions can overstate the significance of predictability to investors in real time (e.g., Goyal and Welch, 2008), we assess out-of-sample predictability. To incorporate information across horizons, we apply out-ofsample mean-combination forecasts (e.g., Rapach et al., 2010). We find positively and statistically significant out-of-sample R2 s for most MA horizons and the mean forecasts. For comparison, we test whether Bitcoin returns are predictable by the VIX, Treasury bill rate, term spread, and the default spread, which are common predictors of stock returns. These variables fail to predict Bitcoin returns both in- and out-of-sample. To assess the economic significance of Bitcoin-return predictability to investors, we form a trading strategy that goes long Bitcoin when the price is above the MA, and long cash otherwise. We find that these trading strategies significantly outperform the buy-and-hold benchmark, increasing Sharpe ratios by 0.2 to 0.6 per year from 1.8. The alphas and mean-variance utility gains are also large for most MA horizons. Moreover, average returns on Bitcoin on days when the MA signals indicate a long Bitcoin position are at least 18 times as large as those when the signals indicate investment in cash. These results are similar across both halves of the sample. The same strategies with various MA horizons also outperform the buy-and-hold benchmark when applied to other two other cryptocurrencies, Ripple and Ethereum, Bitcoin’s two largest competitors. While our model is motivated by Bitcoin, it should also apply to other assets that lack easily interpreted fundamentals.3 Hence, to further test our model, we consider the NASDAQ portfolio during a ten-year window (1996–2005) that includes the dot.com boom-and-bust of the early 2000’s. In this period, many emerging technologies associated with the internet and other communication advances introduced fundamentals that at the time were difficult to assess, much like those of Bitcoin. Many NASDAQ companies possessed no earnings (or negative earnings) for years before they became viable, and other company’s innovations never proved valuable and eventually failed. We show that our technical trading strategies applied to the NASDAQ outperform the buy-andhold benchmark in this ten-year window. This outperformance largely derives from avoiding the length and severity of the major NASDAQ drawdowns during this period. For similar reasons as those motivating our NASDAQ analysis, our model should also apply to individual young internet stocks during the tech boom before they had relatively informative fundamentals. During the fiveyear window around the peak of the tech boom (1998–2002), we find that the average MA strategy increases Sharpe ratios for almost all stocks in the Morgan Stanley Internet Index, which included many representative tech firms of the time (e.g., EBay and Amazon). These results demonstrate wider applicability of our model to other emerging assets characterized by fundamentals that are difficult to value besides Bitcoin and other cryptocurrencies. Our model provides a testable economic implication in addition to the predictability of Bitcoin returns by MAs of prices. Specifically, in our model, trading results from variation across MA horizon indicators. Consistent with this implication, we show that proxies for disagreement across horizons and total turnover implied by the MA signals are significantly and positively associated with trading volume. Hence, overall, our results demonstrate that Bitcoin returns are predictable by MAs of different horizons, investors can profit from this predictability, and Bitcoin’s trading volume is explained by differing MA trading signals across horizons 1.1. Related Literature Our paper contributes to the growing literature on the economics of cryptocurrencies and the associated blockchain technology. Unlike our paper, relatively few papers in this vein study the asset pricing properties of Bitcoin. Using the Cagan model of hyperinflation, Jermann (2018) empirically examines the relative contribution of shocks to volume and velocity on variation in Bitcoin’s price. Jermann finds that most of the variation in Bitcoin’s price is attributable to volume shocks, consistent with stochastic adoption dominating technology innovations. Dwyer (2015) explains how cryptocurrencies can have positive value given limited supply. Athey et al. (2016), Bolt and van Oordt (2016), and Pagnotta and Buraschi (2018) all provide models in which the value of cryptocurrencies depends on some combination of (i) usage and the degree of adoption, (ii) the scarcity of Bitcoin, and (iii) the value of anonymity. Our model differs from those used by prior studies in at least two important respects. First, our model does not require Bitcoin to be interpreted as a currency per se. We do not directly specify currency-related determinants of its value (e.g. (i)–(iii) above). Rather, we model the flow of utilityproviding benefits as a random state variable, which we call a “convenience yield”, but admits a more general interpretation. This generality is important because some market participants argue that Bitcoin is better thought of as a speculative asset than a currency (e.g., Yermack, 2013). For example Bitcoin’s high volatility eliminates its use a store of value, a defining feature of money. Second, the papers cited above all assume full-information, however, our model features learning. This feature is critical given the lack of agreement on what determines the value of Bitcoin.4 The learning aspect of our model also helps us to answer novel questions relative to the prior studies such as: what predicts Bitcoin returns? Relative to asset pricing inquiries, such as ours, most of the literature on the economics of Bitcoin seeks to identify problems, implementation issues, and uses of cryptocurrencies. B¨ohme et al. (2015) discuss the virtual currency’s potential to disrupt existing payment systems and perhaps even monetary systems. Harvey (2017) describes immense possibilities for the future for Bitcoin and its underlying blockchain technology. Balvers and McDonald (2018) describe conditions and practical steps necessary for using blockchain technology as a global currency. Easley et al.
http://apps.olin.wustl.edu/faculty/liuh/Papers/Bitcoin_DLSZZ19.pdf
bitcoin accepted here!