U.S. Housing Starts Through 2001: A Forecast Using

Lagged Housing Starts and the Industrial Production Index

 

CHRIS LASSITER

College of Business

Western Carolina University

 

Abstract

 

Privately owned housing starts in the U.S. are expected to fall seventy percent throughout the next two years.  This forecast is based on the Keynesian model of aggregate expenditures, assuming total housing starts are determined by two-year-lagged housing starts and the Index of Industrial Production.  The forecast suggests that potential buyers in the market for new homes will decrease due to an overall reduction in total income as well as increasing interest rates. This lack of confidence will bring about a decline in overall development and construction, and firms will move away from investment and other capital expenditures in order to focus on research and possible future opportunities.  The Fed’s recent moves to increase interest rates to control the expansion of the economy will lead to an overall decline in investment, slowing the American economy. Along with this, coincident indicators such as new building permits and the industrial production index are expected to fall. (JEL: E47)

 

Part 1.  Introduction

 

This paper forecasts privately-owned housing starts in the U.S. from the beginning of the year 2000 until December of 2001.  The explanatory variables are lagged housing starts (HOUST), and the Index of Industrial Production (INDPRO).  The Industrial Production Index is also lagged two years, which gives some insight into the effects of overall GDP on housing starts.  The forecast is based on the Keynesian model of aggregate expenditures, which in this case uses the industrial production index as a proxy for GDP. The sample period for the forecast begins in April 1971 and ends in December 1999. 

 

Housing starts are a significant factor in determining the direction of the economy.  As total income rises and interest rates remain low, U.S. housing starts would be expected to rise, as we have seen in the past few years. In contrast, a forecast showing a decrease in total housing units started would be significant in explaining a slowing economy.  Generally, housing starts fluctuate at the same time as real output changes.  The forecast horizon of two years was chosen because there are many external factors, such as a sharp increase in construction costs due to a shortage of supplies, that could contribute to a decline in housing starts but not be represented in the forecast. 

 

The remainder of the paper proceeds as follows: Part 2. explains the data used to forecast total housing starts; Part 3. describes the economic and theoretical basis behind the Keynesian model of aggregate expenditures used in the forecast; Part 4. presents the forecasting results for the years 2000 and 2001; Part 5. reveals the economic significance of the study; and Part 6. discusses the conclusions for economic policy.

 

Part 2.  Data

 

There were three variables used in the forecast, all of which were taken from the Federal Bank of St. Louis Federal Reserve Economic Data (FRED).  The proxy for real GDP, the Index of Industrial Production, is shown in the database by INDPRO, given in seasonally adjusted annual rates (SAAR) and recorded in monthly intervals.  Housing starts are represented by total new privately owned housing units started, shown in the database as HOUST, and are seasonally adjusted annual rates shown monthly as well.  Other variables used are the 30-year mortgage rate, used to describe the effects of interest rates on housing starts.  The mortgage rate, shown as MORTG in the FRED database, was eliminated from the final forecasting calculations.  Because all data is given in monthly increments, no data transformations are performed.

 

There were three different regressions made during the study.  The first showed the effects of the industrial production index and the 30-year mortgage rate on total housing starts.  The independent variables were not lagged, and the regression was run simply to get some idea of the effectiveness of the variables in explaining housing starts, using a sample range of twenty-eight years and eight months (4/71 – 12/99).  The R-square in this regression was low, at twenty-one percent.  However, through a hypothesis test it can be seen that both the mortgage rate and the index of industrial production are successful in rejecting the null hypothesis, with P-values of 5.02523E-15 and 6.21749E-12.  In the second regression, the data were lagged over the two-year forecast period, and the third regression was run without the 30-year mortgage rate, leading to an R-square of six percent.

 

The third regression is used to forecast housing starts for the years 2000 and 2001.  It is interesting to note that the residuals from the regression showed a group of outliers beginning in September of 1979 through May of 1980; possibly a reflection of the recession beginning at that time due to the oil crisis. 

 

Part 3.  Using the Keynesian Model

Of Aggregate Expenditures In Forecasting

Privately Owned New Housing Starts

 

This forecast assumes the interest rate will remain constant over the next two years and will have no impact on real GDP.  Total new privately owned housing starts are calculated through the sample period of April 1971 to December 1999.  The combinations of lagged housing starts with the lagged Index of Industrial Production are used to forecast housing starts two years into the future. 

 

Because the data studied covered the years from 1971 until 1999, it is important to take into consideration the fact that recently, housing starts have been much greater than they ever have been.  Due to this situation, the forecast may not be as accurate as it would be if, as trends have shown in the past, housing starts in the past few months were decreasing from an increase in interest rates.  Housing starts, on average, have historically been very sensitive to changes in the economy.  It has been shown that as real GDP rises in one year, the next year usually shows steady to increasing housing starts.  Just the same, as consumers predict a good year ahead, their confidence grows and housing starts have been affected positively as a leading indicator.  The specific variables that have been chosen are important factors in explaining why new housing starts act as they do.  In Keynesian concepts, aggregate expenditures are affected by investment.  In relation, housing starts are explained by factors that affect GDP, such as the interest rate and total income.  The forecast of two years gives a good representation of how the dependant variable would act if the economy were held at a constant state over those years.  The forecast calculations are given in monthly periods as well and are based on the Keynesian model for aggregate expenditures, which is as follows:

 

Y = f (I)  (1.

 

The equation used to forecast housing starts shows total new housing starts as a function of lagged housing starts and the industrial production index.  In this equation, Ht represents housing starts, Ht-2 is the symbol for lagged housing starts, the two representing 2 years, and Yt-2 symbolizes the measure of real GDP (the Index of Industrial Production). The final equation is shown here:

 

Housing Starts = f (Ht-2, Yt-2),  (2.

 

Part 4.  Journey into the Future:

 A 2001 Outlook into U.S. Housing Starts

 

The model shown by equation 2 was estimated by ordinary least-squares regression over the sample period of April of 1973 until December 1999. In this regression, the independent variables are lagged two years.  The regression equation is as follows (with t-statistics in parenthesis):

 

Ht = 2062.7071(15.23) – 0.1071(-2.24) Ht-2 – 4.6157(-4.49) Yt-2

 

The intercept of this equation shows that if the independent variables were both zero, ceteris paribus, then the total new housing starts for a given year would be around 2062 thousand units per month.  The R Square for this equation is extremely low, at 6%, which tells us that the independent variables are effective in explaining about 6% of the total variation in the total new housing starts data.  Both coefficients are negative, representing the fact that as the lagged data increases, there is an overall decline in housing starts.   The results of the regression are shown in Table 1 below:

 

 

 

Table 1

Regression Estimates of Equation 2

 

Explanatory Variable

Estimated Coefficient

t-ratio

Intercept

2062.71

15.23

LHOUST

-0.107

-2.24

LINDPRO

-4.616

-4.49

R2 = 0.061

Standard Error = 297.334

Observations = 321

 

 

The forecasts were calculated over the horizon of two years, giving estimates for the years 2000 and 2001. The forecast uses lagged housing starts and the Index of Industrial Production in explaining total new privately owned housing starts.  The forecast presented data on the basis of over twenty-five years of economic records. The results of the forecast are presented in Table 2 below:

 

Table 2

Forecasts of Privately-Owned U.S. Housing Starts for 2000-2001

Month

Forecast

2000.01

1295.32

2000.02

1289.62

2000.03

1289.84

2000.04

1290.10

2000.05

1286.93

2000.06

1280.08

2000.07

1274.86

2000.08

1273.23

2000.09

1276.86

2000.10

1260.03

2000.11

1267.45

2000.12

1253.20

2001.01

1250.76

2001.02

1255.72

2001.03

1252.88

2001.04

1270.23

2001.05

1257.38

2001.06

1264.74

2001.07

1246.19

2001.08

1249.50

2001.09

1251.09

2001.10

1245.65

2001.11

1240.70

2001.12

1228.77

 

This forecast is consistent with the past behavior of the variables.  Although the forecasts are considerably low, with the past several months of the data being extremely high, it is expected that the forecasts would be lower.  There are two possible explanations for this.  Because of the variation in the housing starts data, the lag of two years might be too long, leading to inaccurate forecasts.  The second explanation is that the dependant variable is not responding well to the other variables.

 

 

Part 5.  Forecast Implications: Look Out Below

 

The forecast results are lower than they should be, but they still hold some significance. In December of 1999 the total new housing starts were 1748 (thousands of units).  This yields a decrease of 74% from December of 1999 to January of 2000.  As the past data shows, housing starts have never decreased by this measure at any point in history, but due to the conditions of the past few years, this drop is not unreasonable.  Because in no other year have housing starts been as high as they are in 1996 - 1999, combined with the fact that the forecast took into consideration all months through April of 1971, it would be expected that the forecasts would be low.  This also relates to GDP as well.  Due to increasing interest rates, in the near future it is expected that income will begin to level off.  Along with the increases in interest rates and construction costs, new housing starts are expected to slow down, and there will be decreases from the record-setting months of the past few years.  So in a sense, the forecasts can be seen as both favorable, as related to recent trends in the economy, and unfavorable in measures of accuracy.

 

The forecast does predict the slowing of the economy.  However, housing starts are more than likely not going to fall by seventy-four percent in one month.  Interest rates have begun to increase, and with this, we would expect housing starts to decrease, but not by the magnitude of this forecast.

 

Part 6.  Policy Conclusions

 

Privately owned housing starts in the U.S. are projected to decrease by seventy percent over the next two years.  With increasing interest rates, we should expect the Index of Industrial Production to begin slowing, reflecting a slight decrease in real GDP through 2001.  The slowing economy will be best seen in the job market, where available jobs will continue to decrease, leading to an overall decline in personal income.  This, in turn, depletes the market of potential buyers, lowering investment and justifying the fall in housing starts. 

 

Industries that will be directly affected by this decline in growth are the development, construction and real estate sectors.  However, the fall should in no way be as drastic as this forecast has shown.  In order to prepare for this, firms should begin to invest in research in an attempt to find new markets and future opportunities. 

 

Because the economy has been in such a boom phase over the past few years, no one is really sure what t expect.  However, the government and the Fed are taking precautionary measures in order to control this phase and, hopefully, extend it further into the future.  In relating this to the present economy, it is better to begin turning the boat a safe distance from the shore, instead of running full-steam ahead and charging straight into it, resulting in high inflation, increasing interest rates, low productivity and increasing debt. 

 

Reference

 

Federal Reserve Bank of St. Louis, Federal Reserve Economic Data (FRED) (12-31-98).

http://www.stls.frd.org.fred/. (2-29-00)