A forecast of economic activity, whether it is a forecast of gross domestic product (GDP), or a specific industry, must begin with a good analysis of the overall economy. The general economic environment sets the tone for the more detailed forecasts that are made for specific industries, such as the sales of durable goods and automobile companies. In the longer term, such forecasts must also take into account the effects of population growth and trends, such as the well-known argument by Thomas Malthus that bare subsistence will eventually prevail if the food supply increases only arithmetically whereas the population grows exponentially.
Although economic theory provides the basic framework for a forecast, judgment often plays an important part in making forecasts as well. A good forecaster will decide that particular current circumstances demand that a model that produces a standard result be modified in some way. This is especially necessary when some event outside the usual run of events has a clear economic effect, such as the collapse in oil prices that occurred in 1987.
The literature on time series methods and non-linear models that are used to produce economic forecasts is vast, with innovations being proposed at a rapid pace. The results of many recent studies have found that the use of non-linear models may improve the accuracy of a forecast compared to frameworks that only consider the average linear relationship between a series and other variables. For example, Chauvet and Su (2013) employ a model with three Markov switching processes to represent business cycles as well as structural breaks or outliers in a series.