Regression Tool instruction page
Instruction Regression Tool
Social Media dataset for testing social media signals (Updated data as of 10-2-2018) 10 EUR
The tool is spreadsheet based (Excel) and automatically pulls returns series for a specific crypto from the web (cryptocompare.com). These series then can be used to test various strategies for statistical significance. Main method used in this is a simple linear regression for those somewhat familiar with statistics. More specifically I use a lagged regression. More info on that here https://en.wikipedia.org/wiki/Linear_regression and here Lagged independent variables.
This lagged regression lets you know if a variable is able to predict returns of tomorrow (or any other number of days in the future) with some statistical significance.
For example it tells you that if Twitter recipients of GNT tweets spike today, what will the price most likely do 3 days from now, and is this result just random noise or a relevant indicator.
The spreadsheet allows you to test three strategies;
- Momentum – This tests if the price of today has a relationship with the price of the same crypto tomorrow. Eg, it tells you that if today ETH is up, will it ETH be up again tomorrow or not. Data comes from cryptocompare website and is free. However the API allows a certain maximum queries, so dont query too often.
- Other crypto currencies – This tests if the price of one crypto currency has a leading relationship with another cryptocurrency. Eg, it tells if BTC is up today, will LTC be likely up or down tomorrow. Data comes from cryptocompare website and is free. However the API allows a certain maximum queries, so dont query too often.
- AltcoinAnalytics Twitter – With the tool I developed I keep track of certain social media metrics with regard to cryptocurrency. One of these are Twitter metrics, which can be used to test if they have a leading relationship. Eg, it tells when the total number of Tweet recipients of GNT increases today, what is the relationship with the price tomorrow. This strategy uses data from a csv file shared at AltcoinAnalytics.com. The current csv file is free to obtain there, however since running the queries costs me time and processing power I plan on asking a small fee for this data in the future. The tool can also work without this data (running the other strategies) and that is free to use.
Test set: 20180115 AltcoinAnalytics TW raw free
Latest dataset for social media metrics can be purchased with the link below (10 EUR, data is from 10 feb 2018 onwards):
There is no such thing as certainty in financial markets. Though a strategy can have some statistical basis that does necessarily mean it will always hold. Moreover, there are always regime shifts or structural changes that are never caught in in statistical models. The foundation in this (as with all statistics) are historical relations, which do not necessarily have to hold in the future.
Additionally also beware of datamining. With a significance level of 5% there is a 1 in 20 chance of finding a false positive. By just randomly trying data there every 1 in 20 relation found (with 5% significance) is a result of randomness. Use this tool as an aid and not crystal ball.
An instruction on how to use the tool can be found in the sheet. For the fetching of data from cryptocompare the powerquery addin needs to be installed. One can see from the information in the sheet (bottom left) if it is likey that Powerquery is installed.