Summary
- We conducted a multivariate analysis of the price of bitcoin with financial data from early 2019 to mid-2020.
- Our main conclusion is that bitcoin should do well in what we could call a “nervous bull market”.
- In this scenario, we would see the market generally going up, with some expectation of inflation in the future, all of this against a bearish backdrop.
The analysis
We used WarpPLS () to create several second-order indices (as composites of first-order index funds) and link them in an exploratory model to help us understand what has been driving the price of bitcoin from early 2019 to mid-2020.
The period from early 2019 to mid-2020 was used because prior to it bitcoin was generally perceived as a cash-like currency that could be used for day-to-day transactions among individuals and organizations. From early 2019 onwards, the perception shifted to one of a store of value; something akin to “digital gold”.
We collected and analyzed daily data from various funds. More specifically, the price of one share of each fund at each day’s close was used. In terms of WarpPLS settings, the outer model analysis algorithm used was “PLS Regression”, and the default inner model analysis algorithm was “Linear”. The composite variables were made up of the following funds.
- FIN, reflecting a bullish view of financial institutions, was made up of the iShares U.S. Regional Banks ETF (IAT), and the Financial Select Sector SPDR Fund (XLF).
- HDG, reflecting a bearish view of the market (intention to hedge), was made up of the iShares Silver Trust (SLV), the SPDR Gold Shares (GLD), and the iShares 20+ Year Treasury Bond ETF (TLT).
- MKT, reflecting a bullish view of the market, was made up of the SPDR S&P 500 ETF Trust (SPY), and the Invesco QQQ Trust (QQQ).
- GBTC, reflecting the price of bitcoin, was measured through a single indicator, namely the Grayscale Bitcoin Trust (GBTC).
The Grayscale Bitcoin Trust (GBTC) provides one of the most straightforward ways for investors to own bitcoin. It is generally available to retail investors through various online brokers.
The results
The figure below shows our model with the main results. The iGBTC variable is an instrumental variable that controls for the effect of “time” on the results; to account for autoregression, or the fact that the variable GBTC is influenced by its own values back in time. The instrument used was a numeric variable generated based on the date associated with each data point. In a previous analysis published on this blog, based on the same data, we did not employ this type of control, which led to slightly different results.
The path coefficients (indicated as beta coefficients) reflect the strength of the relationships; they are a bit like standard univariate (or Pearson) correlation coefficients, except that they take into consideration multivariate relationships (they control for competing effects). A positive beta means that an increase in a variable is associated with an increase in the variable that it points to.
The P values indicate the statistical significance of the relationship; a P lower than 0.05 means a significant relationship (95 percent or higher likelihood that the relationship is “real”). The R-squared value reflects the percentage of explained variance for the variable in question; the higher it is, the better the model fit with the data.
I should note that the P values have been calculated using a nonparametric technique, which does not require the assumption that the data is normally distributed to be met. This is good, because I checked the data, and it does not look like it is normally distributed.
The results
The figure below shows our model with the main results. The iGBTC variable is an instrumental variable that controls for the effect of “time” on the results; to account for autoregression, or the fact that the variable GBTC is influenced by its own values back in time. The instrument used was a numeric variable generated based on the date associated with each data point. In a previous analysis published on this blog, based on the same data, we did not employ this type of control, which led to slightly different results.
The path coefficients (indicated as beta coefficients) reflect the strength of the relationships; they are a bit like standard univariate (or Pearson) correlation coefficients, except that they take into consideration multivariate relationships (they control for competing effects). A positive beta means that an increase in a variable is associated with an increase in the variable that it points to.
The P values indicate the statistical significance of the relationship; a P lower than 0.05 means a significant relationship (95 percent or higher likelihood that the relationship is “real”). The R-squared value reflects the percentage of explained variance for the variable in question; the higher it is, the better the model fit with the data.
I should note that the P values have been calculated using a nonparametric technique, which does not require the assumption that the data is normally distributed to be met. This is good, because I checked the data, and it does not look like it is normally distributed.
So, what does the model above tell us? It tells us that:
- As a bullish view of financial institutions (FIN) increases, the price of bitcoin (GBTC) also increases, in a statistically significant way (beta=0.13; P below .01). This is not normally what one would expect, if we assume that bitcoin’s success means the failure of financial institutions.
- As a bearish view of the market (HDG) increases, the price of bitcoin (GBTC) also increases, in a statistically significant way (beta=0.42; P below .01). This is what one would expect, if we assume that bitcoin is used as a hedge against a drop in the market. Note that this effect is the strongest in the model, by far.
- As a bullish view of the market (MKT) increases, the price of bitcoin (GBTC) also increases, in a statistically significant way (beta=0.17; P below .01). Again, this is not normally what one would expect, if we assume that bitcoin’s success means that a bear market is under way.
The three predictors above (i.e., FIN, HDG, and MKT) explain 33 percent of the variance in the variable GBTC (R-squared=0.33). This essentially means that the model is incomplete, although it does explain enough of the variance in GBTC to be useful in an exploration of major influences on the price of bitcoin.
Main conclusion
While the results above may look contradictory, they in fact suggest that bitcoin should do well in what we could call a “nervous bull market”. Here we would see the market generally going up, with some expectation of inflation in the future (which tends to be good for financials), all of this against a generally bearish backdrop.
Disclosure - As a bearish view of the market (HDG) increases, the price of bitcoin (GBTC) also increases, in a statistically significant way (beta=0.42; P below .01). This is what one would expect, if we assume that bitcoin is used as a hedge against a drop in the market. Note that this effect is the strongest in the model, by far.
- As a bullish view of the market (MKT) increases, the price of bitcoin (GBTC) also increases, in a statistically significant way (beta=0.17; P below .01). Again, this is not normally what one would expect, if we assume that bitcoin’s success means that a bear market is under way.
The three predictors above (i.e., FIN, HDG, and MKT) explain 33 percent of the variance in the variable GBTC (R-squared=0.33). This essentially means that the model is incomplete, although it does explain enough of the variance in GBTC to be useful in an exploration of major influences on the price of bitcoin.
Main conclusion
While the results above may look contradictory, they in fact suggest that bitcoin should do well in what we could call a “nervous bull market”. Here we would see the market generally going up, with some expectation of inflation in the future (which tends to be good for financials), all of this against a generally bearish backdrop.
The author does not own bitcoin at the time of this writing.
