TL;DR
This paper introduces CRIX, a novel dynamic index for the cryptocurrency market designed to adapt to its extreme volatility and rapid structural changes. By employing model selection criteria like AIC, CRIX aims to provide a representative and investable benchmark that goes beyond simple market capitalization weighting, offering a more nuanced view of the market's performance.
What Problem Does This Paper Address?
The cryptocurrency market presents a unique challenge for traditional index construction. Its extreme volatility, the constant emergence and disappearance of new cryptocurrencies (altcoins), and the rapid shifts in market structure make it difficult to create a stable and representative index. Existing indices often rely on a fixed number of constituents, which quickly become outdated in such a dynamic environment. The authors aim to address this by developing a methodology that can dynamically adapt to these changes, ensuring the index remains a relevant and accurate reflection of the market. This is crucial for economic analysis, investment strategies, and the development of financial products like ETFs.
Methods
The core of the paper lies in the proposed methodology for constructing the CRIX index. Key techniques include:
- Dynamic Index Construction: Unlike static indices, CRIX dynamically adjusts its constituents and weights. This is achieved through a model selection process.
- Akaike Information Criterion (AIC): AIC is used as the primary model selection criterion to determine the optimal number of constituents for the index. This criterion balances model fit with complexity, penalizing indices with too many constituents.
- Top-Down Approach for Constituent Selection: Assets are initially selected based on market capitalization (for CRIX) or trading volume (for LCRIX) and then evaluated for their contribution to the index's performance.
- Nonparametric Density Estimation: To implement AIC, the authors use nonparametric methods, specifically an Epanechnikov kernel, to estimate the density of log returns.
- Weighting Schemes: Two primary weighting schemes are proposed:
- Market Capitalization Weighting: Similar to traditional indices, but with dynamic constituent selection.
- Liquidity (Trading Volume) Weighting: This scheme gives more weight to cryptocurrencies with higher trading volumes relative to their market capitalization, aiming to capture market focus and potentially reduce Bitcoin's dominance.
- Monthly Rebalancing and Quarterly Review: The index constituents are reviewed monthly to capture market momentum, with the number of constituents reviewed quarterly.
Datasets
The paper utilizes cryptocurrency data from a proprietary RDC CC database, kindly provided by CoinGecko. This dataset includes daily closing data for prices, market volume, and market capitalization in USD for over 1000 cryptocurrencies. The primary time period for the analysis is from April 1, 2014, to March 25, 2017. The methodology is also tested on German stock market data (DAX, MDAX, SDAX, TecDAX) and Mexican stock market data (IPC35).
Key Insights and Ideas
- Bitcoin's Dominance is Not Absolute: While Bitcoin dominates the cryptocurrency market in terms of market capitalization and trading volume, it does not account for all market variance. Other altcoins play a significant role in market movements, and their inclusion is critical for accurate tracking.
- Liquidity Matters: The paper highlights that trading volume can be a more indicative measure of market focus than market capitalization alone. A liquidity-weighted index (LCRIX) shows a different performance profile, often giving more weight to altcoins and potentially offering a more representative view of market sentiment.
- Dynamic Adaptation is Essential: The extreme volatility and rapid evolution of the crypto market necessitate a dynamic approach to index construction. Static indices quickly become obsolete.
- AIC as an Effective Selection Tool: The AIC criterion proves to be a robust method for selecting the optimal number of constituents, balancing accuracy with parsimony. It often results in sparser indices compared to other criteria.
- Investability is Key: The authors emphasize the importance of constructing an "investable" index, meaning it should be composed of liquid assets that are relatively easy to trade.
Key Results
- CRIX Performance: CRIX, constructed using AIC and market capitalization weighting, demonstrates strong performance in tracking the cryptocurrency market, outperforming Bitcoin in certain periods, especially when altcoins experienced significant gains.
- LCRIX Performance: The liquidity-weighted index, LCRIX, shows a distinct performance profile, often giving higher weights to altcoins, which can lead to different market insights compared to market-cap weighted indices.
- Sparsity and Performance: The proposed dynamic methodology, particularly with AIC, often results in sparser indices (fewer constituents) while maintaining comparable or even superior performance in terms of Mean Squared Error (MSE) and Mean Directional Accuracy (MDA) compared to other model selection criteria.
- German and Mexican Stock Markets: The CRIX methodology was successfully applied to traditional stock markets (Germany and Mexico), yielding more accurate benchmarks (FDAX and FIPC) than existing indices (DAX and IPC) in terms of MDA and MSE, demonstrating the generalizability of the approach.
Limitations and Future Work
- Data Availability and Quality: The paper acknowledges that missing data is observed in the dataset and describes how it is handled (e.g., Last-Observation-Carried-Forward). However, significant gaps or consecutive missing data points can lead to the exclusion of time series, which might affect index representativeness.
- "Local Optimum" for CRIX: The search for the optimal number of constituents for CRIX using AIC might terminate at a local optimum rather than a global one, although the authors argue that this still yields asymptotically optimal results and good performance.
- Future Work: The authors suggest exploring further variants of the CRIX family and applying the methodology to other dynamic markets. The potential for creating ETFs or contingent claims based on CRIX is also mentioned. The paper also hints at further analysis of the theoretical underpinnings of the AIC's effectiveness in this context.
Summary updated