A “Newstrace” analysis example with pinalyze.com

In previous posts, I described a method for finding trading signals in online news articles called “Newstrace” with some tools and Python scripts and a new service I called “Pinalyze” which helps to apply this method. In the post “News history for stock prediction” I showed the features of Pinalyze.

In this article, I take one news article and show how to apply the method and the system. On April 22. in 2019 “Financial Express” published the article:

The article is about billboards which are getting smarter and connected to deliver dynamic and interactive content.

Pinalyze found similar articles published on 29 different days back until 2016–02–24

Similar news in the past

The similar news was published by “Forbes”, “Reuters”, “Yahoo” and others and are about digital outdoor advertising. For example:

and so on.

Pinalyze found that Facebook (FB) is the only one of the 209 analyzed companies which daily changes of stock prices are statistically significantly different on the 29 publishing days compared to the other days in the same timespan.

The difference between the mean of the daily changes of FB / Facebook on and off the “Newstrace” is -2.327. This indicates a movement down of Facebooks price.

But is this difference large enough to show a signal?

The difference could be by random with no systematic effect behind. Pinalyze checks with statistical tests the probability that this result comes only by choice, and found a probability of 4.9% for that. In statistics usually, the hypothesis that it is only a random effect in rejected if the value is lower than 5%. This is the case in our example and so with an error probability of lower than 5%, we can expect some nonrandom effect.

It seems that a small systematic movement of Facebooks stocks took place on days with such news like this in the past.

Pinalyze shows the estimated probability distribution of the changes on the “Newstrace” compared to the rest. The distribution is slightly shifted to the left with the mean shifted by a larger amount than the whole distribution due to some outliers with changes about — 40.

Density estimate for daily changes

Let us look a little bit closer to this aspect by checking the histogram:

Histogram of daily changes in Facebook stock

We see there is an outlier which is also visible in the boxplot:

Boxplot of the daily changes

Looking at the “Newstrace” line chart of Pinalyze you can see that the drop of about 40 points was on July 26. in 2018 with high transaction volume.

On this day the “Forbes” article

were published.

  • Could it have some impact on the drop?
  • Or are there some other reasons?

If we exclude this news from the statistical analysis the significance level goes up above the 5% level and we would not assume a signal.

In a future article, I will show another example of a signal and analysis.