Luka Jankovic & William Nuelle, August 2019
Summary. In this research insight, we deconstruct bitcoin’s performance during periods of macroeconomic uncertainty over the past decade, providing an analysis of historical correlations, market conditions, return clustering, and macro events.We find that:
- Historically bitcoin’s correlation among established macro assets typically hovers within ±0.25 around a 0 correlation; however, we see that bitcoin’s correlations reach their global maxima and minima, that is their most correlated, in Q3 2019, potentially suggesting that bitcoin is becoming part of investors’ reaction functions. In particular, we see stronger positive correlations with fixed income, gold, and JPY (traditionally safe haven assets) and negative correlations with aggregate global equity indices during Q3 2019.
- An analysis of stratified market conditions suggest that bitcoin’s return magnitude is higher when the global market moves are also significant; bitcoin’s median 1-day forward returns are inconsistently positive and negative during significant macro moves, so it’s difficult to generalize and conclude in aggregate whether bitcoin acts as a hedge or a source of liquidity during macro uncertainty.
- Bitcoin maintains its uncorrelated nature during periods of significant market moves better than gold, fixed income, and the yen; US fixed income remains negatively correlated with equities, the dollar, and crude, and positively correlated with other risk-off assets such as global fixed income, gold, and the yen, suggesting it may better offset losses (or gains) in other parts of the portfolio during large market moves.
- While we are limited in the inferences we can draw because of the imperfect mapping between cluster centers and specific macroeconomic environments, a k-means clustering analysis suggests some general environmental “ambiences” where bitcoin performs well (most macro assets are negative, or gold is positive) and poorly (equities are very negative, while bonds are positive).
- Our event-based analysis suggests that in aggregate larger events increase the magnitude of bitcoin’s Cumulative Abnormal Returns (CAR); bitcoin responds with greater magnitude to foreign debt, trade, and debt ceiling events, and with lesser magnitude to environmental and geopolitical events.
Bitcoin on the global stage
Emerging macro trends
One persistent narrative across the digital asset ecosystem is that bitcoin has grown from a niche experiment into an important uncorrelated, macroeconomic asset. Many believe that bitcoin can be used as a portfolio hedge against macroeconomic uncertainty or as a flight to safety under stress conditions, while others contend that bitcoin (and crypto assets broadly) sits at the higher end of the risk spectrum and consequently act as a source of liquidity during periods of portfolio stress. In practice, individual investor behavior and micro reactions to market events are widely heterogenous, but teasing out key aggregate reaction functions of the broader market under macroeconomic uncertainty can have both tactical and strategic implications for portfolio strategy. In order to understand the role that bitcoin can potentially serve in global macro portfolios under periods of macroeconomic uncertainty, we performed multiple analyses that highlight correlation, forward returns, clustering, and event-based analysis dynamics.
Global correlations over time
As bitcoin’s network, value, and portfolio usage has grown over time, we’d expect to see bitcoin’s correlation to interconnected macroeconomic assets increase in magnitude over time. To conduct this analysis, we evaluated bitcoin against a basket of relevant global assets and indices from July 19, 2010 to July 29, 2019 .
The charts above suggest there is no apparent long term “growing” trend for bitcoin’s correlation among more established macroeconomic assets, and typically hover within ±0.25 around a 0 correlation. However, one notable feature of these correlation intervals is that bitcoin’s correlations reach their global maxima and minima, that is their most correlated, in Q3 2019 for 7 of 16 assets. In particular, we see stronger positive correlations with fixed income, gold, and JPY (traditionally safe haven assets) and negative correlations with aggregate global equity markets during Q3 2019. Because Q3 is not yet complete, we can’t test for statistical significance, but current indications suggest it’s possible that recent developments mark a new chapter in bitcoin’s correlation with the global market.
BTC forward returns against daily macro performance
The above finding prompts a follow-up question: even though historically bitcoin has been an uncorrelated asset, are there specific conditions during which bitcoin responds to global market conditions? One place to look is during extreme conditions — both when markets are in strife and when markets are elated. We ran the analyses across all indices and assets and have selected the few outputs that produced interesting results. The x-limits represent intervals of all days with a given daily index performance, and y-limits at 1-day forward bitcoin returns.
We can also observe this dynamic very clearly in times of large market moves, when bitcoin has a greater correlation with established assets, though still relatively low. Since the BTC median 1-day forward returns are inconsistent during significant macro moves, it’s difficult to conclude in aggregate whether bitcoin acts as a hedge or a source of liquidity during macro uncertainty.
We use the S&P 500 as a proxy for market behavior, and compare macro correlations of traditionally uncorrelated, risk-off assets: bitcoin, gold, yen, and US fixed income (Fig. 3). We see that bitcoin maintains its uncorrelated nature during periods of significant market moves better than gold, US fixed income, and the yen. In addition, US fixed income remains negatively correlated with equities, the dollar, and crude, and positively correlated with other risk-off assets such as global fixed income, gold, and the yen, suggesting it may better offset losses (or gains) in other parts of the portfolio during large market moves.
Response to specific macro paradigms using clustering
Clustering algorithms transform large undifferentiated datasets of arbitrary dimensionality into sets of points that are grouped geometrically (for our analysis, we chose the k-means clustering algorithm). Because we can’t visualize our 16-dimensional dataset, we will briefly share a simple example of the method limited to only BTC versus S&P performance in the plots below. The left plot shows a two-dimensional dataset of bitcoin daily performance against S&P 500 daily performance, and the right plot illustrates the data is transformed into clusters.
We clustered our 16-dimensional space into 9 clusters. The center of each return cluster (akin to the colored groupings in the example above) is presented below, along with the resultant bitcoin performance on that day.
Each row can be understood as the center of a specific “region” in space, where each region encodes a macroeconomic environment.  It’s not a perfect translation from cluster centers to comprehensible macro environments because the algorithm forces the centers to be the mean of the nearest points and thus fungible and not tethered to a specific historical example. However, some sense can still be made of the general “ambiance” of each cluster: for example, Cluster 1 is a region where most assets are down across the board and bitcoin performs well. Cluster 4 can be thought of as a region where equities perform very poorly, bonds perform well, and bitcoin underperforms. Cluster 8 is notable because it’s the region where gold outperforms and bitcoin performs the best.
Beyond these qualitative intuitions, we are limited in the inferences we can draw because of the imperfect mapping between cluster centers and specific macroeconomic environments. At the same time, the results are also remarkable: bitcoin’s daily performance was not used as a component in each vector, and the differentiation in its performance across clusters comes entirely as a result of the quality of the regions themselves. In other words, bitcoin is clearly responding differently to different macroeconomic situations, though we can’t say with any meaningful precision what these situations are. Clustering offers intriguing analytical potential moving forward.
Another particularly topical question is “how does bitcoin respond to events, not environments?” Studies of events have an important place in economic literature.  Accordingly, we created a dataset of global events that had macroeconomic implications between January 2010 and July 2019.  In total, nearly 500 observations are included with density of observations increasing up to the present data. The tumultuous 16-day stretch from May 2019 is shown below as an example.
In all, the events were labelled based on their event type, which included the labels Trade Event, Foreign Debt Event, Central Bank Rate Change Event, Geopolitical Event, Corporate Event, Environment Event, Domestic Event, Debt Ceiling Event and New Legislation Event. California-related Events were also included because some portion of the event data set was sourced from California economic records but these were sequestered in the analysis.
Case study: May 2019 China-US trade escalation
Some events may affect bitcoin’s movements more than others. From the events excerpted above, one can hypothesize that China’s tariff announcements will have a larger impact on bitcoin’s price than the Uber IPO. As such, this escalation of trade war may provide a fertile ground to exemplify a generalizable method for the study of all macroeconomic events during bitcoin’s short history.
The value of an event can be studied by estimating Abnormal Returns (MacKinlay 1997) at time t, given in literature as:
To generate the expected normal return, we use a Constant Mean Return Model with normal returns that are given by the historical structure of bitcoin drawdowns: our expected returns are conditioned by bitcoin’s position in its global trend (see footnote for references and description of model development and assumptions). Cumulative Abnormal Return (CAR) measures the abnormal return in a window to either side of the event and is given by:
Here our window size is 7 days, which allows time for events to be priced in. Upon running the analysis, the largest 1-day change in CAR, 0.157, occurs from t=0 to t=1 . We see a 0.174 CAR across the entire timespan. The results conform to our expectations that Day 0 will yield the largest e (abnormal return) during our entire event cycle, especially in the case where the event is an unexpected announcement.
Evaluating all event observations using this methodology sheds a light on the events that are macroeconomic agitators for bitcoin and on events that fail to attract the attention of the market. The chart below shows median CAR values across each unique Size-Type classification of events with a 7-day window:
In aggregate, we see that the size of the event increases the magnitude of the CAR. Looking at the events, we see bitcoin responds with greater magnitude to foreign debt, trade, and debt ceiling events, and with lesser magnitude to environmental and geopolitical events.
There are some clear limitations to this method, though. For example, the model cannot distinguish between two events on the same day — it merely records the CAR for a day on which an event takes place and groups events alike in type. With enough observations of each event type, the accumulation of events creates an informative picture of response impulses, but low counts in any type can easily cause anomalies.
Post-event forward return correlations
Finally, we wish to evaluate bitcoin’s correlation with assets that have well-studied relationships with the global macroeconomic environment in the short timespan after relevant global events. We’ve seen above that bitcoin has historically been a largely uncorrelated asset that appears to respond strongly to specific set of narrow market conditions. A final inquiry into the correlations between forward returns was attempted to study how bitcoin relates to other assets during these specific, narrow market conditions.
The correlations divided by event type produce some fairly noisy, inconsistent results, and it comes as no surprise. We also analyzed a random walk time series and performed the same correlation. The random walk had the same mean correlation per event across all observations, meaning it is highly unlikely this method of analysis is producing useful results. One obvious culprit is that the sample sizes were likely not robust enough in many cases once events were split into specific groups to perform correlations. Another culprit could be the difficulty correlating forward returns between bitcoin, whose market never closes, and other assets which do not trade on weekends and holidays. Therefore, a forward return analysis could be interrupted by a weekend time frame, when bitcoin trades and other assets do not.
We believe that bitcoin remains a largely uncorrelated asset, but it’s clear from results outlined in this research that bitcoin is responding to specific and narrow sets of conditions, and in sometimes inconsistent ways. It appears major events and extreme rises and drops impact bitcoin, but smaller events and less pronounced conditions agitate less, suggesting that bitcoin may still be nascent relative to the global market.
Notes & references
 Bloomberg LP [S&P 500 Total Return Index, STOXX Europe 600 Hedged USD Net Total Return, Nikkei 225 Total Return USD Hedged Index, MSCI Emerging Markets Net Total Return USD Index, Bloomberg Barclays US Agg Total Return Value Unhedged USD, Bloomberg Barclays Pan-European Aggregate Total Return Index Value Hedged EUR, Bloomberg Barclays Asian Pacific Aggregate Total Return Index Value Hedged USD, Bloomberg Barclays EM USD Aggregate Total Return Index Value Unhedged, Bloomberg Dollar Spot Index, EUR/USD, USD/JPY, GBP/USD, CNH/USD, Deutsche Bank Emerging Market Currencies Basket Index, Gold United States Dollar Spot, WTI Crude Oil], 2019, Coinmetrics, 2019.
 As a note, bitcoin’s daily returns across its entire history are normally distributed with mean 0.5% and standard deviation 5.6%.
 MacKinlay, A. Craig. “Event Studies in Economics and Finance.” Journal of Economic Literature. Vol. XXXV. March 2017, pp. 13–39.
 State of California, Department of Finance, Chronology of Significant Events, http://www.dof.ca.gov/Forecasting/Economics/Chrono_Sig_Events/, 2019.
 There are a handful of methods for determining in general cases including the Constant Mean Return Model, Market Model, Factor Model, and Capital Asset Pricing Model. The absence of economic features in bitcoin from which expected return models for Equities like the Market Model, Fama-French Models can be derived means that bitcoin is limited to baseline statistical expected return models like the Constant Mean Return Model, which describes with being the mean daily BTC return. It’s obvious that this method will fail for bitcoin, as it has no recourse to the underlying structure of bitcoin’s price history and will invariably mis-value future expected returns. However, we do have some recourse with the historical structure of bitcoin price movements and their expected returns from our prior work Galaxy Digital Research: Contextualized Analysis of Bitcoin Drawdowns (August 2019). Specifically, we base our expected returns on the probability distributions from our prior drawdown research which approximate a Gaussian distribution as more observations are collected.
Legal Disclosure: This document, and the information contained herein, has been provided to you by Galaxy Digital Holdings LP and its affiliates (“Galaxy Digital”) solely for informational purposes. This document may not be reproduced or redistributed in whole or in part, in any format, without the express written approval of Galaxy Digital. Neither the information, nor any opinion contained in this document, constitutes an offer to buy or sell, or a solicitation of an offer to buy or sell, any advisory services, securities, futures, options or other financial instruments or to participate in any advisory services or trading strategy. Nothing contained in this document constitutes investment, legal or tax advice. You should make your own investigations and evaluations of the information herein. Any decisions based on information contained in this document are the sole responsibility of the reader. Certain statements in this document reflect Galaxy Digital’s views, estimates, opinions or predictions (which may be based on proprietary models and assumptions, including, in particular, Galaxy Digital’s views on the current and future market for certain digital assets), and there is no guarantee that these views, estimates, opinions or predictions are currently accurate or that they will be ultimately realized. To the extent these assumptions or models are not correct or circumstances change, the actual performance may vary substantially from, and be less than, the estimates included herein. None of Galaxy Digital nor any of its affiliates, shareholders, partners, members, directors, officers, management, employees or representatives makes any representation or warranty, express or implied, as to the accuracy or completeness of any of the information or any other information (whether communicated in written or oral form) transmitted or made available to you. Each of the aforementioned parties expressly disclaims any and all liability relating to or resulting from the use of this information. Certain information contained herein (including financial information) has been obtained from published and non-published sources. Such information has not been independently verified by Galaxy Digital and, Galaxy Digital, does not assume responsibility for the accuracy of such information. Affiliates of Galaxy Digital own investments in some of the digital assets and protocols discussed in this document, including Bitcoin. Except where otherwise indicated, the information in this document is based on matters as they exist as of the date of preparation and not as of any future date, and will not be updated or otherwise revised to reflect information that subsequently becomes available, or circumstances existing or changes occurring after the date hereof. For all inquiries, please email firstname.lastname@example.org. ©Copyright Galaxy Digital Holdings LP 2019. All rights reserved.