What's wrong with all the quantitative valuation models we see today in cryptoland?
Humble advice from someone who got destroyed again and again by the market in his 8 years of quantitative trading career:
Humble advice from someone who got destroyed again and again by the market in his 8 years of quantitative trading career:
1a. Do not base investment decisions on models that lack any rigorous statistical evidence, no matter how intuitive these models are.
Charts that show that a model has successfully predicted past 3 bubbles are *not* rigorous statistical evidence.
Charts that show that a model has successfully predicted past 3 bubbles are *not* rigorous statistical evidence.
1b. AFAIK I was the first one to publish a statistical analysis of NVT. Yet people have used NVT to make market calls long before.
medium.com/@QwQiao/nvt-an…
And even then I don't think my analysis is rigorous enough for NVT to be used. It's merely a pointer for future research.
medium.com/@QwQiao/nvt-an…
And even then I don't think my analysis is rigorous enough for NVT to be used. It's merely a pointer for future research.
2. Many empirical papers don't make enough effort to describe input variables and data source. When you read those papers, be skeptical of the result, because "garbage in, garbage out".
As a matter of fact, my own blog post on NVT suffers from this problem.
As a matter of fact, my own blog post on NVT suffers from this problem.
3. Many models are overfitted, i.e., tweaked to perform well in past dataset but fail in future. Rule of thumb: the fancier a model looks the more likely it's overfitted.
It takes years of data science experience to appreciate the statement "complexity is a form of laziness".
It takes years of data science experience to appreciate the statement "complexity is a form of laziness".
4. Beware of models built from pure data mining and lacking theoretical foundation. One such example is models that predict "BTC will go to X because mining cost is Y".
It takes years of experience to appreciate the statement "best regularization method is domain knowledge".
It takes years of experience to appreciate the statement "best regularization method is domain knowledge".
5. More importantly I think crafting quantitative models to automate trading is not the most productive way to approach this market. Quantitative models work only if patterns persist, but because crypto moves at 10000 mph, regime shifts occur too often for patterns to persist.
