Managing cryptocurrency risk through Realized Variance forecasting
One of our day-to-day activities at Cloudwall is to build risk management tools for institutional investors. One of the most commonly used and simplest post-trade risk measures is Realized Variance (RV). In line with our most recent exploration of cryptocurrency stylized facts, today we’ll review a paper focused on RV forecasting and risk decomposition into RV and jumps.
The conclusion of this paper is interesting as it has different implications for long- and short-term investors in digital assets.
The paper begins with “Understanding and managing the risk of the cryptocurrency market is crucial”, which is our mission at Cloudwall, so great start already. The authors notice right away that at such an early stage, this new asset class can be extremely volatile and present frequent price jumps (we have noticed over the past two weeks). BTC is maturing, so we don’t see the same 5 minute jumps anymore, but in this 5-minute sample dating 2017–2020 the price changes range between -19% and +10%. Wild. If compared to traditional finance where these kind of jumps happen rarely, in crypto at this stage of their life cycle we can’t afford not to price that jump risk for the purpose of risk management.
As an example, BTC has 64% of days entangled with jumps, while for the most liquid stocks in S&P 500 this is 8% (and similar in the developed FX market).
The data is from the Gemini exchange, but they also get them from Poloniex, Bittrex and Bitfinex in order to construct a smoothed-out series. It’s not surprising that the jumps are entangled in the 64% of the days for the Gemini series vs 39% of the days for the smoothed cross-exchange series. Does this answer the question why investors should diversity their holding on several exchanges?
The authors separate their price jump estimator from RV and show that this jump estimator is biased due to the presence auto-correlation in jumps. This bias introduces a “false negative” problem, but we can mitigate it with a threshold mechanism to smooth consecutive jumps, based on the Threshold BiPower Variance (TBPV) estimator. Next, they decompose RV into up- and down-side risk, which allows them to explore the effect of jump asymmetries between up and down price moves.
Next, we move to forecasting with HAR (Heterogenous AutoRegression) models, used in financial research to model volatility at different time horizons, recognizing heterogeneity across traders and investors based on their forecasting horizon. The authors show that there’s a significant temporal dependence between 1-day lagged realized variance and jump estimators on the future 1, 7 and 30 days RV. For the decomposed risk, only the positive jumps have a significant (negative) impact on 7 and 30 day risk. They also show that the decomposed risk (with signed jumps) is a significant predictor of longer term RVs. As we could expect, there’s a temporal structure in the considered risk measures, as it evolves systematically in the three years of this dataset, which can be due to the presence of structural breaks. More evidence is required for this specific statement.
For the short term modelling (1 day), including only the lagged RV outperforms adding jumps or signed RV. On the other side, with the increase in the horizon, including jumps improves the accuracy significantly. Given these considerations, the conclusion the authors bring forward is that BTC is used not as a transaction medium, but as an alternative investment.
Another interesting consideration is that the price doesn’t seem to be driven mainly by sentiment, but there’s a considerable impact from financial related news, i.e. exogenous shocks.
The methodology that links financial news and jumps is not particularly scientific, as they manually match top 5 jumps to certain events (such as an ETF application rejection by SEC, showing the largest jump on that particular day). Given the number of 2020 market moving even, it’s also quite hard to link jumps to particular news, especially without a properly timestamped events dataset.
Finally, to evaluate the economic impact of RV forecasting they use the Realized Utility (RU) framework, which mimics a trading strategy where an investor targets a constant Sharpe ratio and adjusts her positions based on RV forecasts. The authors suggest that short-horizon investors should not account for separated jump components for their forecasting models, while long-term investors (30 days horizon) including more predictors helped to increase their investor utility.
What’s our takeaway? Risks are there, diversification helps and dynamic model re-calibration is a necessity, not a good to have.
Reference: Hu, J., Härdle, W. K., and Kuo, W., “Risk of Bitcoin Market: Volatility, Jumps, and Forecasts”, 2019. Source: https://arxiv.org/abs/1912.05228
Disclaimer
Not a financial advice, solicitation, or sale of any investment product. The information provided to you is for illustrative purposes and is not binding on Cloudwall Capital. This does not constitute financial advice or form any recommendation, or solicitation to purchase any financial product. The information should not be relied upon as a replacement from your financial advisor. You should seek advice from your independent financial advisor at all times. We do not assume any fiduciary responsibility or liability for any consequences financial or otherwise arising from the reliance on such information.
You may view this for information purposes only. Copy, distribution, or reproduction of all or any portion of this article without explicit written consent from Cloudwal is not allowed.
