Degree Project in Economics of Innovation and Growth, Second Cycle, 30 credits Stockholm, Sweden 2018 Analysis of Cryptocurrency Market and Drivers of the Bitcoin Price Understanding the price drivers of Bitcoin under speculative environment Yasar Kaya
52
Embed
Analysis of Cryptocurrency Market and Drivers of the Bitcoin Price …1295584/... · 2019. 3. 12. · Analysis of Cryptocurrency Market and Drivers of the Bitcoin Price Understanding
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Degree Project in Economics of Innovation and Growth, Second Cycle, 30 credits
Stockholm, Sweden 2018
Analysis of Cryptocurrency
Market and Drivers of the Bitcoin
Price
Understanding the price drivers of Bitcoin
under speculative environment
Yasar Kaya
Analysis of Cryptocurrency Market and
Drivers of the Bitcoin Price
Yasar Kaya
Master of Science Thesis INDEK 2018, TRITA – ITM – EX 2018:582
KTH Industrial Engineering and Management
SE-100 44 STOCKHOLM
Master of Science Thesis TRITA – ITM – EX 2018:582
Analysis of Cryptocurrency Market and Drivers of the Bitcoin Price:
Understanding the price drivers of Bitcoin under speculative environment
Yasar Kaya
Approved
2018-07-03
Examiner
Hans Lööf
Superviosor
Ulrika Stavlöt
Abstract In this paper, the price fluctuations of Bitcoin under speculative environment is studied. It
has been seen that the market trend points out an existence of a speculative bubble. Over
the course of the period from 2014 to 2018, the trend in price movements of bitcoin has
proved to be strongly speculative. In that regard, investors might be curious about what
drivers might be instrumental in these speculative price changes. After reviewing of NPV, it
was seen that NPV is not applicable to the case of cryptocurrencies due to their nature and
lack of free cash flows to base the asset valuation to some fundamental facts. Later, LPPL
model is reviewed, however, that also proved to be insufficient since it does not reflect the
investor speculations and inform much about price dynamics regarding behavioral finance
principles. Then, some papers from the past price fluctuations of bitcoin (for the period from
2010 to 2013) was reviewed and three key variables were determined which might explain
price movements. Public interest towards Bitcoin as interest-driven, regulatory and political
news about cryptocurrencies as event-driven and VIX as overall investor approach to Bitcoin
market have been taken. After running regressions, the only significant variable happened to
be public interest and popularity of Bitcoin. Although, for some cases, VIX variable also
explain price fluctuations for some intervals, in none of the cases event-driven variable has
long- terms effect on price fluctuations under speculative environment. Lastly, a robustness
test is also handled considering the “weekend effect” and it has been seen public interest
variable again proved to be a significant price determinant.
Before starting to interpret the results of this regression, one might ask if natural logarithm
of public interest variable and the natural logarithm of prices are in line. The correlation
between these two yielded 0.94 which proposes at times there is an increase in the natural
logarithm of prices, the natural logarithm of public interest variable also increases and via
versa.
The results of this regression can be seen in table 4 below. As it is seen, its R Square value is
0.92 which is higher. The initial price of bitcoin labeled as "intercept" is significant and 4.86$.
The only significant variable is GoTrend parameter variable with the confidence level of %95.
Also, %1 increase in GoTrend leads to a %1.18 increase in bitcoin prices.
34
One reason why VIX index variable is not significant in this model in table 3 might be
attributed to the fact that its correlation with bitcoin prices was too low until December
2016 (as it is shown in detail in the part 3.2.2). So, another regression starting from the date
December 4th, 2016 to May 4th, 2018 can be run to see if it will have a significant effect.
However, as it is seen from the table 5, for this period, none of the variables are significant.
Table 3 Regression Results
Table 4 Logarithmic Regression Results
35
Table 5 Logarithmic Regression Results
3.4 Robustness Test
In this part, a robustness test is run by choosing closing prices of Mondays for bitcoin instead
of Fridays. When the same regression is run for Monday, it is compared with the results of
those of Fridays. As it is seen from the table 6 below, GoTrend is again significant with a %95
confidence level. In this case, one unit of increase in GoTrend variable leads to 301$ increase
in prices. Unlike table 3’s results, VIX variable is not a significant variable for the case of
Mondays.
As it is seen in table 7 below, in logarithmic OLS regression, “Intercept” and GoTrend are
significant with confidence levels of %95. The initial price of bitcoin is 3.24$ and %1 increase
in GoTrend leads to a %0.8 increase in prices.
Considering the second period when VIX correlates with bitcoin prices negatively (as of
5.12.2016), for this period of data in table 8, none of the variables happened to be significant
when Monday closing weekly prices were taken.
36
Table 6, Regression for Robustness Test for Mondays
Table 7, Logarithmic Regression for Robustness Test for Mondays
37
Table 8, Logarithmic Regression for Robustness Test for Mondays
3.5 Durbin-Watson Test for Testing Autocorrelation
In the stock market, throughout the history, autoregressive models which capture dynamic
effects over time-varying process have been used. This comes under the premise “past
values of the dependent variable have an effect on current values of dependent variable”. If
the effect of this premise proved, the inclusion of lagged dependent variable in the model is
needed to capture dynamic effect. One might think if this is also the case for Bitcoin market
and price movements (Chan & Chu, 2017). To decide whether to include lagged dependent
variable in the model, autocorrelation test was undertaken for the general model (table 3).
One of the ways for confirming the presence or absence of autocorrelation is handled by
using Durbin-Watson test. In DW-test, if the test statistic is close to 2, the test result
suggests there is no autocorrelation. If it is close to 0, it suggests a positive autocorrelation
whereas if it is close 4, it suggests a negative autocorrelation (Armstrong, 2012). In the first
general model (table 3), the p-value is 0.045 which points out the test is significant. Also, the
Durbin-Watson test has yielded the test statistic of 2.4 which rules out the existence of
autocorrelation for bitcoin prices in this study’s dataset. According to the interpretation of
this value in Durbin-Watson test, the hypothesis H1 which is “the autocorrelation exists” can
be rejected for the first-order autocorrelation. Then, H0 which is “there is no autocorrelation
among residuals” cannot be rejected. So, the residuals are not systematic and there is no
38
need for autocorrelation correction in the model. In this case, the model does not include
the lagged dependent variable (Durbin & Watson, 1971).
Figure 16, the scatter plot of residuals over time for having an idea about autocorrelation Source: own figure
In addition to DW-test, the scatter plot of residuals in figure 16 above, over time (time series
data of residual terms) also suggests no autocorrelation. As it is seen, until the date index of
160, the values hover around the straight zero line. After the time index 160, there is not a
systematic pattern; there is either too low or too high scattered values around 0 line,
confirming randomness (Chan & Chu, 2017).
3.6 Discussion of Results
One of the first things to mention is that VIX variable’s correlation to prices happened to be
0.024 which was not something expected in the sense that it must have been negatively
correlated to the bitcoin prices. However, it is a value very close to 0. So, this case might be
ignored, and it is taken as non-correlated to the price (Kmenta, 1986). When it comes to the
second period of analysis (2016-2018), a higher correlation rate which is -0.47 was obtained.
This is in line with the initial expectation that VIX and prices are negatively correlated.
According to the risk/return tradeoff theory, higher VIX values spell with more volatility in
-10000
-8000
-6000
-4000
-2000
0
2000
4000
6000
8000
0 20 40 60 80 100 120 140 160 180 200
Res
idu
als
of
Reg
ress
ion
Mo
del
Time Index of dataset from 1 to 196
39
traditional markets and the investors do not engage in investments in less mature markets
like cryptocurrency market (Campbell, 2005). When it comes to the results of three
regressions for Fridays, the only regression which has yielded significant VIX is the first OLS
regression (non-logarithmic regression for period 2014-2018). For Mondays, none of the
regressions yielded significant results for VIX. In this case, “Weekend Effect” might be kicking
in and deteriorating the consistency in the results. Yet, inconsistent results between
Mondays and Fridays data might be attributed to the bias due to the small dataset (196
observations for the 2014-2018 period, and 60 observations for 2016-2018).
Considering “Weekend Effect”, one derivation might be done through the “intercept” values
in the regressions. For Fridays, the non-logarithmic regression and logarithmic regression for
the period 2014-2018 has yielded significance. In table 3, the value of -879$ was not
sensible. However, when the logarithmic OLS regression was run for ridding the model from
massive scale difference in between variables, a more sensible value of 4.86, in table 4, was
obtained. This case suggests the initial price of bitcoin is 4.86$ when all the independent
variables are not in effect in the model. For Mondays, the logarithmic OLS regression, in
table 7, yielded 3.24$ of initial price which is lower than 4.86$. This case suggests the
influence of “Weekend Effect” which proposes Fridays asset prices in speculative markets
might be higher than Mondays.
For News’ variable, none of the regression tests yielded significant results. This variable was
put into the model to capture the possible event-driven impact of news’ released on bitcoin
prices and volatility on weekly basis. The results are in line with the results of past studies.
Pryzmont (2016) studied for the period 2014-2016 and he also ruled out the effect of
bitcoin-related “incidents”. Hence, it is possible to conclude, on weekly basis, there is no
impact of news on bitcoin prices. One possible explanation for that might be attributed to
the absence of reliable legislations and regulations supported by the banks and
governments. In this case, the investors do not consider the news released as serious as it is
for traditional markets, in the long-run the effect is minimal. However, in the short-run
(hourly basis), sudden price changes due to fear and panic might be observed (Turpin, 2014).
GoTrend variable happened to be the most relevant variable in general for almost all the
analyses. To begin with, its strong correlation of 0.89 with the prices hinted robust
40
expectations in the first place. The logarithmic regressions for Mondays and Fridays give
sensible results (table 4 and table 7). With the “Weekend Effect”, for Fridays, the impact of
interest-driven table GoTrend is higher in prices. Kristoufek (2013) also found similar results
for the period 2013 to 2015. The two periods suggest the same results with some difference.
In the period of Kristoufek’s (2013) study period, the price movements were explained with
the investor excitement and hype in bitcoin.
In summary, the main hypothesis in this paper was checked and different results were
obtained for all three variables. The “Weekend Effect” theory was harnessed to see have a
robustness test. By its nature, this study makes a contribution for bitcoin price analyses
considering “Weekend Effect” and combination of three distinct variables in essence.
41
Chapter 4 ∙ Discussion and Further Research
One of the interesting results in this paper is that there is an inconsistency in results when
Mondays’ and Fridays’ closing prices times series are incorporated into the model. This case
suggests that a more comprehensive study which takes daily closing prices times series into
the model might yield more accurate results. Since all the other data was weekly, for the
sake of consistency, this paper worked with weekly bitcoin closing prices.
Considering the scope of the study, it could have been more informative if the speculative
price movements in the cryptocurrency market were compared with other speculative
bubbles throughout the history. Learnings from other bubbles over the past might
contribute to and enrich the scope of cryptocurrency bubble analysis. Considering the
bubble theory and underlying dynamics, lifetimes of bubbles and speculations from different
industries might pose similar patterns. Since there is a limited time for this study, I could not
afford the time to make such an analysis. Further studies in this respect might engage in
such analysis.
One another point is that the relationship between Bitcoin price movements and public-
interest variable might be bidirectional. According to the findings for public interest
variable’s significance, this paper suggests;
Bitcoin prices do increase as it gets more and more popular throughout the internet and
speculative bubble formation is observed.
However, the case might also be the other way around;
Public interest does increase as bitcoin prices get more and more expensive. In the end, there
is a vicious cycle in which prices increase public interest, and as public interest increases,
prices go higher.
This statement above might open a new research opportunity for the researchers to
understand speculative price movements and bubble formation in cryptocurrency market. In
the paper of Kristoufek (2013), for the period of 2010-2013, the bitcoin prices have been
analyzed and it has been found that there is a strong correlation between bitcoin prices and
GoTrend variable. Also, it was showed that these two variables are bidirectional and
42
influence each other. Later, it was pointed out that this case results from speculation and
“trend-chasing behavior”, in which the bitcoin was the financial trend then.
Lastly, theories about behavioral finance might be studied for better understanding the
speculative environment of bitcoin prices. Then, the findings of this paper might be
harnessed to be explained by considering those theories. Regarding the drivers of
endogenous models, there are various theories to investigate the price behavior such as
animal instinct, herding, bounded rationality, moral hazard, due diligence, predatory
decision making. They might also be considered in the future studies and more far-reaching
conclusions might be drawn.
43
Chapter 5 ∙ Conclusion
In this paper, the price dynamics under a speculative environment of bitcoin is studied. As
there are no government or financials institutions bolstering the bitcoin and its myriads, it is
of a grave challenge to understand why people demand for bitcoin exchange. Due to the
recent popularity of bitcoin and the hype, its value exhibits dramatic fluctuations.
Using the financial data and empirical analyses held in this paper, it was seen that, public
interest has a high correlation. In all regression tests, it has proven to be a significant
variable in influencing the price of bitcoin under a speculative environment of the market.
Only for the case of analysis for the time frame of 2016-2018 for Mondays, it was not a
significant variable. However, this result’s validity is controversial since the limited numbers
of data might draw bias. Considering the “Herd Behavior” and “Greater Fool” theories
together with the hype drawn by the media for bitcoin, speculative price movements and
bubble creation in the market due to an interest-driven variable is not surprising.
In none of the regression tests handled for Mondays and Fridays, the event-driven variable
was significant. This case proposes, on the weekly level, it is not possible to explain
speculative price movements in bitcoin through event-driven effects released in the news.
Considering the VIX, it has shown significance when the model is fed with Fridays’ prices
time series analysis. When the robustness test was held using Monday prices, it has shown
no significance in any of the regressions. This case might be attributed to the “Weekend
Effect” for the longer run empirical analyses. However, considering the lifespan of Bitcoin as
2008 the birthdate and as 2014 the starting date for the speculative environment, it is hard
to claim “Weekend Effect” is influential due to the small dataset.
References
Aurelio F. Bariviera, María José Basgall, Waldo Hasperué, Marcelo Naiouf, Some stylized facts of the Bitcoin market, May 11, 2017
Al Shehhi, A., Oudah, M., & Aung, Z. (2014), " Investigating factors behind choosing a cryptocurrency". In Industrial Engineering and Management, 2014 IEEE International Conference, December 2014 (pp.1443-1447)
Armstrong, J. Scott (2012). "Illusions in Regression Analysis". International Journal of Forecasting (forthcoming).
Baker, Samuel L. (2000). “Perils of the Internal Rate of Return". Retrieved January 12, 2007
Berk, Johnathan; DeMarzo, Peter; Stangeland, David (2015). Corporate Finance (3rd Canadian ed.). Toronto: Pearson Canada. p. 64
Campbell, John Y., and Luis Viceira. The term structure of the risk-return tradeoff. No. w11119. National Bureau of Economic Research, 2005
Chan, Stephen, Jeffrey Chu, Saralees Nadarajah, and Joerg Osterrieder. 2017. “A statistical analysis of cryptocurrencies”. Journal of Risk Financial Management 10.
Cross, Frank. (1973), The behavior of Stock Prices on Fridays and Mondays, Financial Analyst Journal
Dreman, David. (1993). "One More for the Road? " Forbes. New York, 363.
Durbin, J., & Watson, G. S. (1971). Testing for serial correlation in the least squares regression III. Biometrika, 58(1), page: 1–19
Eatwell, J., Milgate, M., & Newman, P. (1987). “The New Palgrave: a Dictionary of Economics”. New York: Palgrave.
ElBahrawy A, Alessandretti L, Kandler A, Pastor-Satorras R, Baronchelli A. “Evolutionary Dynamics of the Cryptocurrency Market”, Royal Society Open Source, 2017.
Fama, E. F. (1965), “The Behavior of Stock-Market Prices”, The Journal of Business, Vol. 38, No. 1, pp. 34-105.
French, Kenneth. (1980), Stock Returns and the Weekend Effect, Journal of Financial Economics
Garber, P., Famous First Bubbles, MIT Press, 2000
Grinberg, Reuben. "Bitcoin: An Innovative Alternative Digital Currency", Hastings Science & Technology Law Journal vol. 4, no. 1 (Winter 2012): p. 159-208.
Johansen, A., Ledoit, O., and Sornette, D. (2000). Crashes as critical points, International Journal of Theoretical and Applied Finance, 3, no. 2: 219-255.
Johansen, A., & Sornette, D. (2010), “Endogenous versus Exogenous Crashes in Financial Markets”, Brussels Economic Review, Vol. 53, No. 2, pp. 201-253
Keynes, John M., “Treatise on Money“ (London: Macmillan, 1930).
Kindelberger, C., “Manias, Panics, and Crashes”, New York, Basic, 1978
Kristoufek, L. (2013, December 4). Bitcoin meets Google Trends and Wikipedia: Quantifying the relationship between phenomena of the Internet era. Scientific reports, 3.
Kmenta, Jan (1986). Elements of Econometrics (Second ed.). New York: Macmillan. pp. 298–334.
Liberatore, Vincenzo. Computational LPPL Fit to Financial Bubbles, January 2011.
MacDonell, Alec, Popping the Bitcoin Bubble: An application of log-periodic power law modeling to digital currency, April 2014.
Marx, Karl. (1867/1887). Capital: A Critique of Political Economy. Vol. 1, The Process of Production of Capital.
Muraki, Tori, XU, http://www.businessinsider.com/bitcoin-price-wall-street-volatility-index-correlation-deutsche-bank-2018-1?r=UK&IR=T&IR=T, January 2018
Pryzmont, Piotr. (2016, August). An Empirical Study of How Bitcoin Related Incidents Impact its Price Volatility. National College of Ireland, Master Thesis
Rodrigue, J-P., Comtois, C., Slack, B. (2009), The Geography of Transport Systems (2nd ed.), London, Routledge, ISBN 978-0-415-48324-7.
Rosser, J. B., Jr, From Catastrophe to Chaos: A General Theory of Economic Discontinuities Kluwer Academic, 2nd ed., 2000
Turpin, J.B. (2014) 'Bitcoin: The Economic Case for a Global Virtual Currency Operating in an Unexplored Legal Framework'. Indiana Journal of Global Legal Studies, 21(1): pp.335-368.
Schumpeter, Joseph A. (1942/2014), Capitalism, socialism, and democracy (2nd ed.), Floyd, VA, Impact Books, ISBN 978-1617208652.
Shiller, Robert. “How High Bitcoin’s Price Can Go in 2018” Interviewed by Robert Hackett, Fortune Magazines, December 21, 2017.
Siegel, J. (2003). What Is an Asset Price Bubble? An Operational Definition. European Financial Management, 9(1), 11–24. Retrieved from http://onlinelibrary.wiley.com/doi/10.1111/1468- 036X.00206/full
Seys, Jen. Deceastecker, Kjartan. (2015-2016). The Evolution of Bitcoin Price Drivers: Moving Towards Stability? University of Ghent, Master Thesis
Sirkin, G., “The Stock Market of 1929 Revisited: a note ‘, Business History Review, Vol. 44, 2 (Summer 1975), pp. 223-231
Sornette, Didier. Critical phenomena in natural sciences: chaos, fractals, self-organization, and disorder: concepts and tools. Springer, 2004.
Sornette, Didier & Cauwels, Peter. Financial Bubbles: Mechanism and Diagnostics. Swiss Finance Institute Research Paper No: 14-28. April 2014
Van Wijk, D. (2013, July 18). What can be expected from the Bitcoin? Erasmus Universiteit Rotterdam.
Victor Chang, Russell Newman, Robert J. Walters, Gary B. Wills. Review of Economic Bubbles, 20 February 2016
Yamey, Basil S., "Scope for Futures Trading and Conditions for Success," in How Commodity Futures Markets Work, Thames Essay No. 42, B. S. Yamey, R. L. Sandor, and B. Hindley, eds. (London: Trade Policy Research Centre, 1985).
Businessdictionary.com (2018), "What is greater fool theory? definition and meaning". Retrieved 2015-03-06.
The Guardian (2018). “Bitcoin biggest bubble in history, says economist who predicted 2008 crash.” [online] Available at https://www.theguardian.com/technology/2018/feb/02/bitcoin-biggest-bubble-in-history-says-economist-who-predicted-2008-crash [Accessed 19 Jun. 2018].