Do Americans feel secure about their jobs ?

Every month, the University of Michigan updates its surveys of consumers by collecting people’s opinion on their current financial situation. One of the questions respondents are asked is:

During the next 5 years, what do you think the chances are that you (or your husband/wife) will lose a job that you wanted to keep ?

On average, the answer is typically around 20%. However, it can evolve rapidly, depending on the political and financial situation in the US:

Because confidence in job security will likely affect a person’s economic behaviors, there is a high incentive to anticipate its evolution in the near feature - for example before making investment or marketing decisions.

How we can predict perceived job security

A first step towards guessing how job security confidence will evolve in the near future is to acknowledge the seasonality of the phenomenon.

As you can see, Americans typically feel less confident about their jobs when the summer hits (and they have to announce to their boss that they are taking an unplanned 2 month break). Also, it is reasonable to assume that their confidence doesn’t change drastically from one month to another. Practically, this means for us is that we can get a good idea of how people are going to feel about their jobs in the next month by using the monthly history of the time series and using an autoregressive model.

In this concrete case, a statistical test shows that the job security confidence was 1 and 6 months before is the most relevant measurements for predicting it at any given time. In order to test how well we are able to make predictions for the future using only these two anterior values, we can compare actual observation to what this model would have predicted:

Not too bad: on average, this model is off only by 1.143 %.

Is there no way to do better than this in the age of data ?

Actually, yes - you can use Google Trends, a website by Google which lets you see how many users make certain queries in any region of interest during a period of your choice. By seeing what people are searching in real time, you can gain a competitive edge and make more reliable predictions of the future evolutions of the market. Coming back to confidence in job security, the idea is very simple. If there is an overnight boom of Google queries for job opportunities, then there must be a growing concern about job security in the general population - no need to wait for the University of Michigan to publish their next report in a few weeks!

To get the most out of Google Trends, you’ll need to carefully think about what queries are relevant. If you cannot read people’s minds, or are lazy, then you’ll benefit from having an automated way of selecting those. After automatically generating a list of candidate queries, we selected the most useful ones with a Bayesian technique called Spike and Slab, which computes estimated probabilities that certain variables should be included in the model.

By using the queries for which inclusion probabilities are high (Loan and Crisis), we obtain a lower mean absolute error (1.1293%):

Baseline MAE Google Trends MAE Improvement of MAE(%)
1.1430 1.1293 1.1970%

While this improvement is not that impressive overall, it appears that Google Trends is very useful in predicting the behavior of the time series in uncertain times, like the Great Recession or the COVID-19 Recession:

This improvement is much more important, because in these periods, timely adaptation is crucial. Since April, due to the instability of the current situation, people are constantly changing their perception of their financial situation. We get much better predictions in this period using Google Trends:

Baseline MAE (COVID-19) Google Trends MAE (COVID-19) Improvement of MAE(%)(COVID-19)
2.266 1.59 29.8239%

What we learned

  1. Google Trends is a simple, publicly-available tool to get insights about the interests of the general population.

  2. Working with Google Trends is not easy as it may look. There are two main restrictions, the result of the query highly depends on the combination of search terms and you can only search for 5 terms at once.

  3. Choice of topics is crucial and not always intuitive.