Final Words

We tested several adaptive MCMC algorithms to estimate the parameters of the \(\text{CAViaR}\) family of models. We found that the Robust Adaptive MCMC algorithm (Vihola 2012) is best suited to explore the target distribution in the case of a heavy-tailed return vector. Given the time taken to run the Monte Carlo simulations, we used C++ to speed up key loops which made it possible to get a forecast in under a minute using a standard personal computer . We were also able to build and deploy a web application using exlusively the R programming language to gather, analyze and visualize real time financial data. Currently, the app depends entirely on its single R process available through shiny, which has serious limitations whenever the server receives multiple messages within a second, which can be pretty intense when markets drop. Since there is only one R process available to treat this incoming data, analyzing it and generating output, the full application can freeze whenever market activity is too high. This issue can currently be handled simply by disabling the WebSocket updates and relying exclusively on HTTP requests happening once per minute instead. However, real time data can be critical for decision-making in which case a possible solution is to delegate this task to the client through Javascript or using a multithreaded web framework to handle the incoming data.

An important portfolio management element has been left aside throughout this document, i.e., the correlation between asset returns. The factor copula models introduced by Krupskii and Joe (2015) offer an intuitive way to model the correlation structure among traditional financial markets and cryptocurrency returns. Another important issue that has been left off this work concerns ensuring non-crossing quantiles as described by Liu and Luger (2017).

Cryptocurrency markets are new and growing exponentially. It is difficult to predict which cryptocurrencies will survive, but it seems like they will all be using some form of blockchain technology for the foreseeable future. This can either mean that organizations will be using this newer and better technology as a replacement to old technology while still relying heavily on human labour or that this newer and better technology will play an important role in the automation of modern labour.