*4.2. Cleaning Trend and Noise Effects in the Cryptocurrency Market* 4.2.1. Noise and Trend

The cryptocurrency market is known to have a higher percentage of noise than other traditional financial markets. According to [66], the average daily signal-to-noise ratio of the cryptocurrency market is 36%, which is extremely low compared to well-established US stock exchanges such as NYSE and NASDAQ, with an average daily signal-to-noise ratio of 90%, given the considered period between March 2017 and November 2017. The noise in the cryptocurrency market might come from different sources. For instance, there is no fixed volume for a transaction to be executed at a time, so investors can freely choose the amount that they want to trade; however, this issue causes one problem, in that investors can reduce the transaction costs by splitting their budget into smaller pieces and then buy one cryptocurrency many times with different amounts of volume and price, a practice which can trigger unforseen price movements, see [67]. Furthermore, cryptocurrencies' prices are vulnerable to "pump and dump" schemes [68], which have become pervasive recently, and also regulatory news enacted by national authorities [69]. All of these factors might intervene in the price movements of digital assets. Consequently, the correlation matrix between cryptocurrencies cannot explain their real connections as it is highly influenced by these noise factors.

On the other hand, the trend effect found in other correlated systems [70] might be found in the cryptocurrency market. Briefly speaking, a trend among cryptocurrencies means that they tend to move together in terms of price values. We notice that the majority of cryptocurrencies are created based on the protocol of leading cryptocurrencies such as Bitcoin and Ethereum (e.g., MKR, BNT, ICX, ETC and LTC) [71]. Moreover, cryptocurrencies' prices readily fluctuate with mass media [72], causing a herding behavior [72]. Similar characteristics contribute to creating a trend in cryptocurrencies.

Generally, these phenomena might be reasons for a high-value correlation matrix of cryptocurrencies from our dataset. Thus, it is important to remove of the existing noise and trend before moving on to further analysis.
