Non-Linear, Linearity for Crypto Price Predictions

Approximately Correct versus Precisely Wrong

As noted by Ray Dalio in his book “Principles”, the current education system has molded decision-making in favor of precision over approximation. He views this mentality as ineffective for deciphering and connecting the meaningful dots from the irrelevant. Similarly, for a variety of academic and societal reasons, financial practitioners and investors have been conditioned to think in the complex and precise rather than the approximate and probabilistic.

Armed with this assertion, we similarly propose approximative methods for crypto price predictions (which we know to be non-linear) using linear models. Yes, of course it is contradictory, but hear me out.

The Power of the Trend

Academics and real-world financial market practitioners both recognize the power of price trends, which has been documented in a bevy of finance literature (a simple Google search will unearth numerous). A fellow crypto fund, Protos Management, discusses trend following and how it can deliver high, risk-adjusted returns here and here. Pugilist has noted that trends are particularly potent in crypto markets given the participant’s propensity to boom / bust psychology.

*data from coindesk.com

Understanding cyclical market trends within the crypto market allows us to unbundle the market into segments or clusters that may be useful for prediction. Clusters typically take three forms:

1. No trend with meandering prices
2. Positive trend with linearly increasing prices (green arrows)
3. Negative trend with linearly decreasing prices (red arrows)

If one can appropriately identify the cluster in which the crypto market is currently within, then one could leverage linear models to predict asset prices, which we know are inherently non-linear.

For example, at the beginning of 2018, it would not have been a far-fetched assertion to presume that a market cluster switch to a negative trend state had occurred, thus one could have made appropriate adjustments to their price prediction models and portfolios. For example, our Historicism for Cryptomarkets was posted back in mid-January stating such, but was met with skepticism by many.

There are three caveats:

1. GIGO — The model is only as good as the variable inputs and user expertise. Thus, if the user does not find good independent variables, overfits their model, or has high multicollinearity, then the results will be poor. The old garbage in, garbage out adage.
2. Lag — There is a decision lag when the market switches clusters which may result in losses until the participant recognizes that a market cluster switch has occurred.
3. Mentality — The participant will need to trust the methodology and endure price gyrations, i.e. the periodic counter-trend price moves (orange arrows) that will ultimately fade away under the pressure of the overall trend (black arrows).

*tradingview.com

Non-Linear, Linearity Theory Examples

Everyone is a genius in hindsight, so lets present some historically, time-stamped examples which lend a little more credence to this theory.

0x Project (ZRX)

Our post, Cryptoasset Valuation with Data Science + Fundamentals was largely ignored and even laughed at back in early March when $ZRX was trading at $0.60, given its simplistic linear methodology. Since then, $ZRX has traded almost exclusively within our predicted price ranges and still does as of July 11, 2018.

Zcash (ZEC)

Back in mid-March, within our full investment report of Zcash (ZEC) (page 12), we made a price prediction when $ZEC was trading around $260, which was recently attained despite intermittent price gyrations. The price movements were largely attributable to the market “end of tax season bottom” and Gemini listing, which ultimately topped out at $382. As mentioned in caveat 3, a stomach for counter-trend price movements is needed until price ultimately succumbs to the overall market trend.

Conclusion

This theory and approach is not re-inventing the wheel and nor is it wholly novel. But, what anyone who has studied data science will assert is that often times, the simplest models are the best ones. Are there limitations to this approach? Absolutely.

However, it does provide a more simplistic framework for crypto market practitioners to analyze and forecast asset prices based upon a common trend following methodology. Furthermore, this approach is more inclusive to retail investors given building a linear model is far simpler than sophisticated machine learning methodologies.

Disclaimer: This analysis has been designed for informational and educational purposes only. Readers are advised to conduct their own independent research into individual assets before making a purchase decision.