# Erik Lindström: What Drives Cryptocurrency Returns? Learning Market Regimes Through the Sparse Statistical Jump Model

**Tid: **
Må 2023-01-30 kl 15.15 - 16.15

**Plats: **
KTH, 3721 (Lindstedtsvägen. 25)

**Medverkande: **
Erik Lindström (Lund)

### Abstract

Hidden Markov models are often used for inferring hidden states in financial markets as these can sometimes be interpreted as market regimes. However, that interpretation implicitly assumes that the underlying state process has a certain level of persistence, which is not always the case, especially for high dimensional models.

We propose a novel estimation approach based on temporal clustering of features by penalizing jumps between clusters. The advantages of the proposed jump estimator include that it learns the hidden state sequence and model parameters simultaneously while providing control over the transition rate, it is less sensitive to initialization, it performs better when the number of states increases, and is robust to misspecified conditional distributions.

Feature selection becomes necessary in high-dimensional settings where the number of features is large compared to the number of observations and/or when the underlying states differ only with respect to a subset of the features. We consequently develop a closed form coordinate descent algorithm for the extended feature selection problem that scales well to large data sets with large numbers of (noisy) features. The usefulness of the proposed framework is demonstrated by comparing it with several other methods, indicating that sparse jump model outperforms all other methods considered and is remarkably robust to noise.

Finally, the framework is applied to five leading cryptocurrencies, where more than 400 features derived from financial markets, sentiment variables and cryptomarkets are considered. The resulting model uses fewer and more persistent states than comparable hidden Markov Models. Furthermore, each state provides a coherent interpretation as a corresponding market regime, indicating the usefulness for applications.