I chose to analyze the stock price of Mediocre Social Network Apps from January 2015 to September 2019. With my analysis, I came up with forecasts of the stock price for the first ten trading days of October. In this blog post, I plan to describe the various components of my analysis and the interpretation of the forecasts!

This is a plot of the stock price over time.

The data looks quite volatile, meaning that there is interchangeable increases and decreases in the stock price. Additionally, there is a clear downward (perhaps quadratic) trend to the data. Finally, we can see that the price varies a little bit more in the first half than the second half of the data. This can be fixed with a simple transformation on the data.

I was able to model the above data using a method known as differencing, which essentially calculates the difference between consecutive data points. Generally, differencing with order “k” (i.e. differencing “k” times) approximately removes a polynomial of degree “k” from the data. I applied this procedure twice in order to remove the seemingly quadratic trend (degree of two) to the data. After differencing twice, this is what the data looked like:

The leftovers after the differencing of the data are known as noise. This noise can be modeled in several different ways, and I was eventually able to narrow it down to a single model from which I produced the following forecasts shown in red:

The first image includes gray bands around the forecasts, which show its uncertainty. They have some uncertainity due to the heavy volatility of the price, but they will keep the company’s expectations realistic. Even from observing the stock price over the last 194 days, it is evident that rapid increases often lead to quick declines, so projecting a slight decrease in the price after growth in the past 20 days is logical.