Labor Market Model Brief

Labor Market Model Brief

Labor Market Model Brief Let’s provide a brief overview of unemployment prediction.

One generally looks at the labor market in terms of unemployment. 

We can also break down the labor market into multiple industries. Unemployment can be seen as an effect of the entire economy. If economic production decreases, e.g., GDP decreases, then we would expect unemployment to decrease as well.

Therefore, predicting unemployment is highly predicated on predicting production. We generally use an ARMA (Autoregressive Moving Average) model to predict GDP, where the GDP in the period we wish to predict depends on past observations.

Additionally, unemployment over the long-term shows the property of mean-reversion.

This means that if we see a deviation from long-term unemployment then we expect that to convert to the long-term mean. For example, during the Great Recession, we saw a high unemployment rate in the construction industry, based on mean-reversion (recovery in the construction industry, more demand for housing) we would see the unemployment rate converge to 5%.

Physics Nobel Prize Winner MIT Prof Frank Wilczek on String Theory, Gravitation, Newton & Big Bang

Putting all of these observations together, we can specify a crude model which estimates unemployment based on the GDP variable and past unemployment using an ARMA(1,1) model.

Here, we’re trying to estimate unemployment next year using the current year’s unemployment and GDP percent growth, since we believe that unemployment will not change too rapidly from past observations and economic production is a crucial input into whether people are employed. Additionally, if we believe that there are further lags, e.g., people may take a while to enter the labor market, then we can add further lags.

Additionally, we can narrow down our prediction to certain industries, noting we should increase our confidence interval as we ask for more precise data.

Here, we are trying to estimate the unemployment rate in a particular industry based on our previous model and any relevant additional variables we can think of. A common method used in public policy is difference-in-difference analysis for certain policies, how would the housing industry be affected by an increase in lumber tariffs, or crude oil price shocks on gasoline refinery profits.

In creating a full labor market model for any industry, one would argue that we would want to include several variables for a multi-year estimation period. In addition to GDP, we should include industry wages, population growth (e.g., growth in college graduates year-over-year), and industry demand (e.g. the oil and gas industry has had lower growth expectations as consumers move to other products such as electric vehicles). These are interrelated and we should then move to a Vector Autoregression model where we make multiple predictions in each time period.

In the above VAR mode, we are predicting unemployment, wages, and demand based on an estimated intercept and the observations in a previous period. It makes sense that as unemployment decreases, wages would increase as employers struggle to hire for instance.

To note, as we add on more variables, we increase the risk of overfitting our model and also increase the complexity of it. The model should be parsimonious as possible, as we want to explain it to investors and colleagues. The Vector Autoregressive Model and the Structural VAR are workhorses of economic and financial predictions, where we take a systems-based approach based on economic logic.

We can use the unemployment prediction as an input into a downstream model. For instance, perhaps unemployment percent levels are significantly correlated with the profits of restaurant delivery services.

Labor Market Model Brief Written by Allen Chen

Using Data to Interpret Homelessness