-
Right charts show most correlated sections of left (total) charts
-
Top two / Bottom two charts show two stocks have low future correlation despite high current correlation
Inspiration
Often, investors use a stock's past performance as a peusdo-metric to estimate future growth. However, it is commonly said that a stock's future growth is not impacted by its past. If two stocks had very low variance in movement between two arbitrary time periods, how much variance in correlation is there between them is the immediately following time periods?
What it does
Simply put, the program sieves through millions of pairs of data segments to find areas of low absolute-variance between the two trends. For each pair, it calculates the absolute-variance sum of the immediately following segment to that of its pair. It also calculates the average variance between all segment pairs between the two stocks. If past movement impacts future growth, the immediately following segment should exhibit a lower-than-average absolute-variance sum.
The results are displayed as a long spreadsheet of percent differences in average data to lowest found data and percent differences of immediately following data to lowest data. The standard deviation of the average differences finds the z-value of the immediately following differences and calculates on a standard normal curve.
How I built it
Correlation statistics is built on Python3. It gets all twenty years of stock information through requests on the AlphaVantage API. For a randomly selected pair of stocks, it finds any possible segment of data in stock1, and for each segment, it finds all possible segments of data in stock2, and for each of these, it calculates an absolute-variance sum of every degree of freedom in the data segment. Using these absolute-variance sums, it can find averages for all iterations per trial, and for all trials (each, using two other randomly selected stocks), values were recorded in a spreadsheet to find deviation on the standard normal distribution.
Challenges I ran into
Stocks split and become half its value in the time of a day. Many companies have not been public for twenty years. Processing all the data took a long time. Sometimes, http requests were not satisfied.
Accomplishments that I'm proud of
I collected a lot of data and drew conclusions from analyzing it. Using distributions to calculate the impact of my data and percent correlation. Incorporating so much of my Python knowledge to finish a project.
Log in or sign up for Devpost to join the conversation.