As cryptocurrencies have exploded in value, so too have the attempts to understand them. Even more exciting is a recent uptick in quantitative analysis. Spreadsheet models have risen in popularity as a tool for evaluating and predicting trends.
The effort and analytic rigor behind these models is phenomenal. However, there are strong reasons why most top funds do not use this as part of their evaluation methodology. Often, there is no objective measure — what am I trying to predict, price in 1 week? 1 year? 10 years? And the feature set is often not very predictive (are we really so sure that the number of Telegram users is a strong indicator of future price?). In this space, it’s all too easy to fall into the trap of cargo cult analysis: building complex models that are not all that predictive once the assumptions and objectives are stated and the model is tested.
To illustrate my point, I’ll walk through an example on an existing dataset and finish with my own take on token sale evaluation. All that’s required to follow along is a bit of basic statistics. I’ll post the code at the end of the article.
Full article, source:
https://hackernoon.com/building-a-good-cryptocurrency-model-is-harder-than-you-think-b4f590edc8e9
