News Sentiment and the Saturday Sandwich Effect

photo

It Was On A Friday

The Inferess Analytics Teams constantly strives to understand the data we base our news analytics on. This allows us to discover signals that lie beyond the obvious and bring an information advantage to our customers. It's a continual process. This Friday was no different in that regard.

Understanding the data often involves exploratory data analysis using visualisations. This process led to us to notice some interesting patterns in the positive versus negative balance of news over a week.

So We Asked a Simple Question

Where does bad news go to hide?

Is there a preferred day?

Could the answer could be found by querying over the sentiment labelled news found in our backtest data?

We spent the rest of the afternoon pulling levers and twiddling dials to extract the answer from several months of Inferess Quantified News.

And We Got An Answer

The answer that materialised was surprising and reproducible over differing calendar ranges of backtest data.

Bad news hides on a Sunday with Friday as a close second choice.

This is illustrated by the graph below. For each day of the week the relative proportion of negative news to positive is showed as stacked bar.

The findings are based on samples of many thousands of news articles1 allowing statistical significance to be attributed to these news sentiment fluctuations.

Which Led to More Questions

Negative news is often regarded as carrying more actionable information than positive news. The Saturday Sandwich Effect rises further questions we explore in later blogs posts,

  • When the news is sliced by event type what other skewed distributions arise?
  • How efficient can a market be when news is released out of trading hours?

We will leave theories accounting for this phenomenon to behavioural investment analysts and other experts but the pattern is clear and likely explanations not hard to propose.

Find Out More About Inferess Quantified News

In the context of seeking trading opportunities or enriching risk models Inferess Quantified News is always watching and measuring for your benefit. If you would like to explore the use of Inferess Quantified News please contact us today and we can start working together to add alternative data into your models.

Notes

1. Study based on 185,637 documents sampled from Inferess Quantified News over a 6 week period in 2016.

FOLLOW US

short