Momentum or Contrarian. Which Is the Most Valid in the Case of Cryptocurrencies?

Transcription

1 Momentum or Contrarian. Which Is the Most Valid in the Case of Cryptocurrencies? Krzysztof Kość Paweł Sakowski Robert Ślepaczuk QFRG & Labyrinth MIMUW / 28

2 Motivation What? Investigate the presence and potential strength of momentum and contrarian effects in the cryptocurrency market Why? Momentum/contrarian effects were identified in the past on young and inefficient markets Cryptocurrency market is young, volatile, and rapidly growing No one has investigated this yet Construct an investment strategy giving abnormal rates of return? How? Construct ranking of TOP00 crypto with the highest market cap Construct momentum/contrarian portfolios Calculate descriptive statistics Benchmark against reference strategies Perform sensitivity analysis of parameters 2 / 28

3 Agenda Briefly about cryptocurrency markets 2 Briefly about momentum/contrarian 3 Hypothesis 4 Methodology 5 Data 6 Results 7 Summary 8 Research extensions 3 / 28

4 Cryptocurrency markets 4 / 28

5 Cryptocurrency markets 5 / 28

6 Momentum and Contrarian effects Momentum/Contrarian - classical anomalies present on young and ineffective markets. Momentum - Tendency for the trends of price changes to continue Contrarian - Tendency for the trends of price changes to reverse 6 / 28

7 Hypothesis Main Hypothesis: The momentum and/or contrarian effects are currently present on the cryptocurrency market. Research Questions: How strong magnitude? 2 Which effect is stronger? 3 Short/medium/long- term? 4 Practical possibility of profit? 7 / 28

8 Methodology - Construction of Ranking During each day: Filter out crypto having 4-day MA volume lower than VF = 00 USD 2 Pick 00 crypto with the largest market cap We arrive with a N days 00 matrix that from now on we will call The TOP00. Note We now can use TOP00 to construct rankings for any ranking intervals RA d. 8 / 28

9 Methodology - Main Parameters %N - the percent of TOP00 assets that will be used in portfolio construction 2 Reallocation period (RE) - distance between two neighbouring reallocation days Reallocation day - the day we update the composition of our investment portfolio based on some kind of ranking (market cap TOP00 in our case). 3 Ranking window (RA) - time interval used in TOP00 In general RA!= RE 4 Transaction costs (TC) - as a percentage of total portfolio value 5 Volume filter (VF) - the threshold value for 4-day MA filter 9 / 28

10 Methodology - Portfolios & Benchmarks We use TOP00 to construct the following portfolios: Momentum - equally-weighted investment in %N = 25% of cryptocurrencies with the highest weekly rate of return, assume RE = w and TC =.0% 2 Contrarian - equally-weighted investment in %N = 25% of cryptocurrencies with the lowest weekly rate of return, assume RE = w and TC =.0% And judge their performance in comparison with the benchmark portfolios: S&P B&H - buy and hold reference investment using the S&P500 index and the same time horizon 2 BTC B&H - buy and hold reference investment using the BTCUSD pair and the same time horizon 3 EqW - equally weighted reference investment in all the assets present on TOP00, assume same parameters RE = w and TC =.0% 4 McW - market cap weighted reference investment in all crypto present on TOP00, assume same parameters RE = w and TC =.0% 0 / 28

11 Methodology - Portfolio Efficiency Using on TOP00, calculate the total gross rate of return: where: T ( R (p) 0,T = + t= N i= N the total number of assets w i,t r i,t W R t TC ), () T is the investment s total time horizon (measured in days) w i,t is the percentage (weight) of the i-th asset in the whole portfolio p on day t r i,t is the simply accruing daily rate of return of the i-th asset on day t W R t is the total portfolio turnover rate (in percent) on day t TC is the total percent transaction costs / 28

12 Methodology - Descriptive Stats To benchmark our strategies we also need: annualised rate of change (ARC): ARC = ( + P ) 365 T T, (2) P 0 2 annualised standard deviation (ASD): ASD = 365 T (r t r) 2, r t = P t (3) T P t t= 3 maximum drawdown coefficient (MDD): ( ) MDD (T ) = max max P t P τ τ [0,T ] t [0,τ] 4 information ratio coefficients (IR, IR2): (4) IR = ARC/ASD IR2 = sign(arc)arc 2 /(ASD MDD) (5) 2 / 28

13 Data Daily OHLC prices, market cap and 24h-volume data 2 In-sample time horizon: to for 200+ cryptocurrencies 3 BTCUSD and S&P500 daily close prices as benchmarks 4 Data source: 3 / 28

14 Data histograms Market Cap Volume (24h) Number of Observations Mean=6,079,98 Min=0 Max=00,438,000,000 Total Incomplete=.6% Total Variables=223 Number of Observations Mean=,446,369 Min=0 Max=4,48,070,000 Total Incomplete=9.05% Total Variables= Data Incompleteness [%] Data incompleteness [%] 4 / 28

15 Data filtering Missing values handling: Close: Fill missing observations with last non-missing entry 2 MarketCap: Calculate missing from the circulating supply approximate formula: MC t = ( + r t ) MC t. 3 Volume: Filter out all observations for which 4-day rolling mean volume < VF = 00 USD After that construct TOP00. 5 / 28

16 Data histograms - TOP00, after cleaning Market Cap Volume (24h) Number of Observations Mean=89,277,869 Min=0 Max=00,438,000,000 Total Incomplete=4.2% Total Variables=450 Number of Observations Mean=2,268,025 Min=0 Max=4,48,070,000 Total Incomplete=0% Total Variables= Data Incompleteness [%] Data Incompleteness [%] 6 / 28

17 Sample of crypto data First 0 cryptocurrencies in TOP00 as of Name %ARC %ASD %MDD IR IR2 Start Date MarketCap [USD] Volume (24h) [USD] %MISS bitcoin ,369,600,000,403,920, ethereum ,40,400, ,424, ripple ,806,200,000 26,864, bitcoin-cash ,83,520,000 78,037, litecoin ,966,700,000 7,063, dash ,52,090,000 47,092, nem, ,78,830,000 4,67, bitconnect Inf ,22.7 Inf ,558,580,000 0,550, neo 2, ,443,000,000 25,368, monero ,327,650,000 25,397, Last 0 cryptocurrencies in TOP00 as of Name %ARC %ASD %MDD IR IR2 Start Date MarketCap [USD] Volume (24h) [USD] %MISS zencash ,749,900,464, edgeless 8, , ,07,500 96, aragon ,87, , rlc ,397,600 23, taas 2, ,407, , nolimitcoin 8, , ,97,600 84, nav-coin ,209, , loopring ,275,700 88, wings 4, ,63, , kin ,996,200 38, Legend: Inf - more than 00,000 Start Date - the first day the asset has appeared on TOP00 %MISS - percentage of missing data 7 / 28

18 Results I: In-sample Name %N RE RA %TC VF %ARC %ASD %MDD IR IR2 %MT S&P B&H BTC B&H McW 00 w EqW 00 w Momentum 25 w w Contrarian 25 w w Legend: McW - MarketCap weighted strategy, EqW - Equally Weighted strategy, %N - percent of TOP00 currencies used to construct the portfolio, RE - reallocation period, RA - width of the ranking window used to calculate the highest/lowest rates of return, %TC - total transaction costs, VF - volume filter threshold, %ARC - annualised rate of return, %ASD - annualised standard deviation, %MDD - maximum drawdown, IR, IR2 - risk-weighted gain coefficients, %MT - portfolio mean turnover ratio. Time horizon: / 28

19 Results II: In-sample Time horizon: EqW portfolio is the most efficient among other benchmarks 2 Strong outperformance of contrarian strategy over reference portfolios 3 Momentum portfolio performs better than reference portfolios from regulated markets being worse than crypto benchmarks 9 / 28

20 Results - In-sample Momentum Momentum, %N=25, RE=w, RA=w, TC=.0%, VF=00 Portfolio Performance 00% 000% EqW Momentum McW BTC B&H S&P B&H Drawdown 80% 40% 0% / 28

21 Results - In-sample Contrarian Contrarian, %N=25, RE=w, RA=w, TC=.0%, VF=00 Portfolio Performance 00% 000% 0000% Contrarian EqW McW BTC B&H S&P B&H Drawdown 70% 40% 0% / 28

22 Sensitivity Analysis - Parameters %N = 5%, 0%, 25%, 50% 2 Reallocation period RE = d, w, m 3 Ranking window RA = d, w, m 4 Transaction costs TC - 0.5%,.0%, 2.0% 5 Volume filter VF = / 28

23 Sensitivity Analysis I: In-sample Benchmark Strategies Name %ARC %ASD %MDD IR IR2 %MT %ARC %ASD %MDD IR IR2 %MT S&P B&H BTC B&H McW EqW Parameters MOMENTUM CONTRARIAN %N RE RA %TC VF %ARC %ASD %MDD IR IR2 %MT %ARC %ASD %MDD IR IR2 %MT 25 d w , , w w m w w d w w w m w w , w w , w w , w w , w w , w w , w w w w w w w w w w d d Inf Inf Inf w w m m / 28

24 Sensitivity Analysis II: In-sample Sensitivity analysis confirms the initial results 2 Results for various parameters reveal substantial volatility 3 Strong monotonic effect in case of the efficiency of contrarian and momentum strategies: Contrarian portfolio increase their efficiency when: RE decreases RA decreases %N decreases Momentum portfolio increase their efficiency when: RE increases RA increases %N increases 24 / 28

25 Portfolio Diversification: In-sample Investigation of the correlation matrix gives us a tip that any investigated cryptocurrency portfolio has a huge diversification potential when combined with regular investment portfolios represented by S&P500 B&H strategy. S&P B&H BTC B&H McW EqW Momentum Contrarian S&P B&H BTC B&H McW EqW Momentum Contrarian Time horizon: / 28

26 Summary Strong contrarian and momentum effect on cryptocurrency market 2 Contrarian is much stronger than Momentum and reference strategies 3 Sensitivity analysis performed for various parameters confirms our initial results 4 Strong monotonic effect in case of efficiency of contrarian and momentum strategies 26 / 28

27 Research extensions Reproduce results on -minute data 2 Repeat calculations for quotes against BTC instead of USD 3 Check the results for larger set of parameters and more conservative transaction costs and liquidity constraints 4 Show the results on out-of-sample data starting from Repeat calculations for only these cryptos which are quoted on the largest and most reputable cryptoexchanges 6 Prepare an on-line interactive version of this research with weekly update of each strategy 27 / 28

28 Thank you! Krzysztof Kość Paweł Sakowski Robert Ślepaczuk The views presented in this text are those of the authors and do not necessarily represent the views of the institutions with which the authors are affiliated 28 / 28