Purely Data-Based Method for Long Range Forecasting of Labor Productivity

摘要

For the two major funds within the Medicare and Social Security system, thetrustees must make 75 year projections of the income and outgo. Estimates of population and labor productivity growth are used to predict increases in taxable wages. The long range labor productivity estimates are currently based on the best judgement of the trustees. The paper attempts to eliminate the judgmental aspect of these long range projections by taking a purely data-based, mechanistic approach based on a number of time series autoregressive statistical diagnostics of labor productivity. Using 100 years of historical data (from 1889 to 1989), the paper finds two distinct trends in labor productivity; one from 1889 to the early 1930's and a second from the early 1930's to 1989. The report concludes that the actuarial projections should be based on the entire period 1933 to present rather than the current 1950 to present. Also, the projections, using the 1933-1989 data, would appropriately be based on the trend in the level of labor productivity rather than on the trend in the log of labor productivity as is now being done. These changes would result in substantially lower projected rates of growth in labor productivity; in the area of 1.1% rather than the current 1.7 to 2.0% and they would prevent overestimating projected revenues.