Market forecasting – 3 ways to find the market dynamics

Market forecasting – 3 ways to find the market dynamics

Read time
5 mins
Published on
April 16, 2018

I often stumble across professionals working closely with the market which explains the dynamics and drivers of the market with a story.

A story with great compassion which captivates and convinces in almost a heroic way. I recently started to dig into what the story was based upon and realized that it might not have been such a good story after all. It turned out that the story often came from a consultant which used the story to sell his latest report in market research and seldom captured the underlying and recurring market dynamics. This insight inspired me to write this article; providing readers with the tools to find the dynamics and the drivers of the market based on data.

CCF plot

The cross-correlation plot visualizes the delayed relationship between two-time series by plotting the correlation value between the two variables while lagging one of them. The positive aspect of the plot is its ability to explain how a relationship between two-time series functions, however as commonly known, correlation can be misleading, as seen when searching for funny correlations on the web.

Granger causality

Granger causality is a hypothesis test for determining whether one time-series is useful in forecasting another, first proposed in 1969 by Clive Granger who was awarded the Nobel price in economics 2003. Straightforward, it identifies if there is any cause-effect relations between two variables by first evaluating if lagged values of x are individually significant according to a T-test and if all of them add explanatory power to the regression according to an F-test.
However, it’s a hypothesis test, which means that the result will only be a yes or a no and not a number indicating how much one affects the other.

Accuracy improvement by added indicator

The last method is rather a process than a named methodology. It focuses on the accuracy or model-fit improvement when adding variables. Depending on the aim of the analysis, it could either be used for data exploration by focusing on changes in model fit to answer questions such as;

  • What describes the current market?
  • What’s the reason for current market value?

Or for predictive analysis by focusing on either the average out-of-sample accuracy or a specific step ahead accuracy, to answer questions such as;

  • What describes the future market development?
  • What drivers will affect the market?

This methodology solves both the “funny correlation” problem and gives you a scaled answer.

If you are interested in how this methodology is applied, I recommend reading our case study

Subscribe to our newsletter

Thanks for joining our newsletter
Oops! Something went wrong while submitting the form.