CRIX Index Formula in Crypto Markets

• CRIX is a systematic quantitative index that tracks the dynamic cryptocurrency market by selecting and weighting a subset of coins.
• It employs both market-cap and liquidity weighting with an AIC-based framework to optimize index constituents and minimize tracking error.
• The methodology ensures statistical representativeness and effective risk management through regular rebalancing and divisor adjustments.

The CRIX (CRyptocurrency IndeX) index provides a systematic, quantitative representation of the cryptocurrency market, accommodating both the market’s high volatility and its frequently changing structure. The CRIX index is constructed to track the overall crypto market using a dynamic selection of coins, robust weighting schemes, and information-criterion-based optimization to minimize tracking error. This framework enables researchers and practitioners to study cryptocurrency market development and construct CC portfolios with sound statistical underpinnings (Trimborn et al., 2020).

1. Definition and Construction of the CRIX Index

At any given time $t$, CRIX comprises $k$ cryptocurrencies. For coin $i$:

• $P_{i,t}$: price (USD) at time $t$,
• $Q_{i,t^-_l}$: quantity (outstanding shares) at the most recent re-weighting or re-start time $t_l^-$,
• $\beta_{i,t^-_l}$: weight-capping adjustment factor at $t_l^-$.

The CRIX level is given by: $$\mathrm{CRIX}_t(k,\beta) = \frac{\sum_{i=1}^k \beta_{i,t^-_l} P_{i,t} Q_{i,t^-_l}}{\mathrm{Divisor}(k,\beta)_{t^-_l}}$$ where the divisor is set so that CRIX assumes a pre-specified value at $t=0$: $$\mathrm{Divisor}(k,\beta)_0 = \frac{\sum_{i=1}^k \beta_{i,0} P_{i,0} Q_{i,0}}{\mathrm{CRIX}_0}$$

Whenever the index constituents or their $\beta$ factors change, the divisor is adjusted to maintain continuity: $$\frac{\sum_{i=1}^{k_1}\beta_{i,t^-_{l-1}} P_{i,t-1} Q_{i,t^-_{l-1}}}{\mathrm{Divisor}(k_1,\beta)_{t^-_{l-1}}} = \frac{\sum_{j=1}^{k_2}\beta_{j,t^-_l} P_{j,t} Q_{j,t^-_l}}{\mathrm{Divisor}(k_2,\beta)_{t^-_l}}$$

Two principal weighting schemes are utilized:

• Market-cap weighting: $\beta_{i,t^-_l} = 1$
• Liquidity weighting ("LCRIX"): $$\displaystyle \beta_{i,t^-_l} = \frac{\mathrm{Vol}_{i,t^-_l}}{P_{i,t^-_l} Q_{i,t^-_l}}$$

2. Information-Criterion-Based Selection of Index Constituents

The number of index constituents $k$ is optimized to enable CRIX to track the total market index (TMI) closely, balancing representativeness with parsimony. The TMI includes all $K$ coins with available prices: $$\mathrm{TMI}_t(K) = \frac{\sum_{i=1}^K P_{i,t} Q_{i,t^-_l}}{\mathrm{Divisor}(K)_{t^-_l}}$$

Define log-returns: $$\varepsilon^{\,\mathrm{TMI}}_t = \ln \mathrm{TMI}_t(K) - \ln \mathrm{TMI}_{t-1}(K),\quad \varepsilon^{\,\mathrm{CRIX}}_t = \ln \mathrm{CRIX}_t(k,\beta) - \ln \mathrm{CRIX}_{t-1}(k,\beta)$$

The goal is to find the smallest $k$ and (if necessary) $\beta$ such that

$$\min_{k,\beta} \|\varepsilon^{\,\mathrm{TMI}} - \varepsilon^{\,\mathrm{CRIX}}\|^2$$

in order to minimize tracking error. Overfitting is mitigated by the Akaike Information Criterion (AIC): $$\mathrm{AIC}\{\widehat{\varepsilon}(k,\beta), s\} = -2 \ln L\{\widehat{\varepsilon}(k,\beta)\} + 2s$$ Here, $s$ denotes the number of free parameters, $L\{\cdot\}$ is the likelihood estimated non-parametrically via the Epanechnikov kernel. For given $k_1$, $\beta=1$ for the first $k_1$ coins, with extra $s$ weights estimated for any universe of size $k_1+s$.

3. Weighting Schemes and Constraints

Weights are determined at each re-weighting: $$w_{i,t_l^-} = \frac{\beta_{i,t_l^-} P_{i,t_l^-} Q_{i,t_l^-}}{\sum_{j=1}^k \beta_{j,t_l^-} P_{j,t_l^-} Q_{j,t_l^-}}$$ implying $\sum_{i=1}^k w_{i,t_l^-}=1$.

Between re-weightings: $$\mathrm{CRIX}_t = \mathrm{CRIX}_{t_l^-} \sum_{i=1}^k w_{i,t_l^-} \frac{P_{i,t}}{P_{i,t_l^-}}$$

Optimization is exclusively over $k$ and, in the case of liquidity weighting, the additional $\beta$ parameters.

4. Rebalancing and Maintenance Procedure

• Constituent selection: At each month-end, all coins are ranked by market capitalization (or volume for LCRIX), with the top $k$ included for the next month.
• Constituent replacement: When coins leave the top $k$, replacements are inserted; returning coins may re-enter.
• Quarterly $k$ re-estimation: On the last trading day of each March, June, September, and December, the AIC-based algorithm uses the previous three months’ returns to select a new $k$ for the next quarter.
• Divisor adjustments: Updated at both monthly roll-dates and quarterly $k$ revisions according to the index continuity condition.
• Handling missing data: For isolated data gaps, “Last Observation Carried Forward” is used; consecutive missing days prompt removal at next rebalancing.

5. Algorithmic Implementation of Constituent Selection

At each quarterly review, the AIC-driven process is specified as follows:

1. Construct the full TMI of size $K$.
2. Set $k=k_1$ (e.g., 5 or 10) and compute CRIX$(k,\beta)$ using the $k$ largest coins.
3. Increment $k \gets k + s$ (common $s=5$); re-compute CRIX$(k,\beta)$, estimating new $\beta$ via maximum-likelihood with kernel density estimation.
4. Compute $$\widehat{\varepsilon}_t = \varepsilon^{\,\mathrm{TMI}}_t - \varepsilon^{\,\mathrm{CRIX}}_t$$, determine AIC$\{\widehat{\varepsilon}(k,\beta), k-k_1\}$.
5. Terminate when AIC increases; select the preceding $k$ as optimal.

6. Performance Metrics and Evaluation

Principal metrics for analyzing CRIX tracking ability relative to TMI or Bitcoin include:

• Mean Squared Error (MSE):

$$\mathrm{MSE} = \frac{1}{N} \sum_{t=1}^N \left(\mathrm{CRIX}_t - \mathrm{TMI}_t\right)^2$$

where $N$ is the number of days or months evaluated.

• Mean Directional Accuracy (MDA):

$$\mathrm{MDA} = \frac{1}{N} \sum_{t=1}^N \mathbf{1}\left\{ \mathrm{sign}(\mathrm{CRIX}_t-\mathrm{CRIX}_{t-1}) = \mathrm{sign}(\mathrm{TMI}_t-\mathrm{TMI}_{t-1}) \right\}$$

representing the fraction of intervals for which CRIX and TMI move in the same direction.

• Tracking Error (TE):

$$\mathrm{TE} = \sqrt{\frac{1}{N-1} \sum_{t=1}^N (\varepsilon^{\,\mathrm{CRIX}}_t - \varepsilon^{\,\mathrm{TMI}}_t)^2}$$

7. Context and Significance

The CRIX index’s design addresses several challenges unique to the cryptocurrency market, including high velocity of market structure changes, frequent coin turnover, and extreme volatility. By employing an AIC-based constituent-selection framework and robust rebalancing with clearly defined weighting and divisor calculation procedures, CRIX maintains statistical representativeness and minimizes tracking error without the need for manual intervention.

Inclusion of altcoins and appropriate weighting—market-cap or liquidity-based—are both empirically demonstrated to reduce tracking errors and improve index representativeness, even as constituent market capitalizations vary substantially. The methodology’s adoption of non-parametric likelihood estimation (Epanechnikov kernel) for AIC calculation is a key feature that enables rapid response to market evolution (Trimborn et al., 2020).

CRIX provides a transparent and statistically sound methodology for constructing cryptocurrency market benchmarks, offering significant utility for both academic research and financial applications, such as portfolio construction and risk monitoring.