Data

The forecast is based on the national income and product accounts (NIPA) data that were released on February 27, 2009.

The Latest Version of the US Model

For purposes of this forecast the US model has been reestimated through 2008:4. These estimates and the complete specification of the model are presented in Appendix A: The US Part of the MCD Model: March 1, 2009, which is an update of Appendix A in Fair (2004).

Beginning with the forecast dated October 31, 2005, a few minor specification changes have been made to the US model from the version in Fair (2004). These are explained in Changes to the US Model Since 2004.

Assumptions Behind the Forecast

At the time of this writing the stimulus bill has been passed. In order to analyse the effects of the stimulus bill, I have made two forecasts: a baseline forecast and then a forecast with the stimulus effects added.

The baseline assumptions are as follows. The table below gives the growth rates that were assumed for the baseline forecast for the key exogenous variables in the US model along with the actual growth rates between 1993:3 and 2008:4.

              Growth Rates (annual rates)

           Baseline
           Forecast              Actual     
          Assumptions         2008:4-1993.3  

TRGH          6.0                  5.9       
COG           2.0                  4.0       
JG            0.0                 -0.7       
TRGS          5.0                  6.0       
TRSH          2.0                  5.6       
COS          -3.0/1.0              3.6       
JS           -2.0/0.0              1.4       

These variables are the main government policy variables in the US model aside from tax rates. TRGH is nominal federal government transfer payments to households, COG is real federal government purchases of goods, JG is federal government civilian employment, TRGS is nominal federal government transfer payments to state and local governments, TRSH is nominal state and local government transfer payments to households, COS is real state and local government purchases of goods, and JS is state and local government employment.

-3.0/1.0 for COS means that COS was assumed to fall at a 3 percent annual rate for the first four quarters, 2009:1--2009:4, and then to grow at an annual rate of 1.0 percent after that. -2.0/0.0 for JS means that JS was assumed to fall at an annual rate of 2.0 percent for the first four quarters and then to remain unchanged after that. All tax rates for the baseline forecast were taken to remain unchanged from their 2008:4 values.

The above assumptions have state and local governments contracting some for 2009 and then returning to normal. For the federal government everything is business as usual---no stimulus, etc. Again, this is for the baseline forecast.

No assumption is needed about monetary policy for the forecast because monetary policy is endogenous. Monetary policy is determined by equation 30, an estimated interest rate reaction function or rule.

The Baseline Forecast

The baseline forecast results can be seen by clicking "Solve" after "The MCD Model," creating a data set, and then going immediately to "Examine the results without solving the model." You can then examine any variable in the model.

Real GDP Growth and the Unemployment Rate: The baseline forecast has real GDP falling by 2.7 percent in 2009:1, 1.7 percent in 2009:2, and 0.9 percent in 2009:3;. It then grows at 0.1 percent in 2009:4 (essentially no change) and then 2.0 percent in 2010:1. (All growth rates in this memo are at annual rates.) The unemployment rate rises to 9.5 percent by 2009:4. It begins falling in the middle of 2010. The jobs variable, JF, shows jobs falling by 3.11 million in 2009.

Inflation: Inflation as measured by the growth of the GDP deflator (GDPD) is predicted to be about 2.5 percent in 2009 and 2.0 percent in 2010. The model, however, has been overpredicting inflation for the past few quarters, and so the current predictions should be interpreted with some caution. (For the past forecasting record of the US model, see The Forecasting Record of the US Model.)

Monetary Policy: The estimated interest rate rule (equation 30) is predicting that the three month bill rate (RS) will be essentially zero in the next two years. It then rises gradually to 1.4 percent by the end of 2012.

Other Variables: The federal government budget deficit, variable SGP, is predicted to be between about $150 and $175 billion (at a quarterly rate) for the next four years. This leads the federal government debt, variable AG, to be $9.128 trillion by the end of 2012, which compares to $5.786 trillion at the end of 2008. Interest payments of the federal government, variable INTG, rise from $58.7 billion at the end of 2008 to $103.3 billion (at a quarterly rate) at the end of 2012.

The U.S. current account deficit, variable SR, is forecast to be between about $150 and $175 billion (at a quarterly rate) in the next two years.

Comments on the Baseline Forecast and Possible Experiments to Run

One of the reasons the model is predicting negative growth in the first half of 2009 is the negative wealth effect from the fall in stock prices and housing prices. There is also a large inventory correction predicted for 2009:1 because of past inventory buildups.

Why is the economy predicted to be no worse? Why no predicted huge decreases in GDP and increases in the unemployment rate? The answer is roughly as follows. There are two equations in the US model that have large negative residuals for the last two quarters, 2008:3 and 2008:4, the nondurable consumption (CN) equation and the durable consumption equation (CD). Otherwise, the residuals for the other 28 equations are all within what would appear normal. The error terms in the CN and CD equations are not assumed to be serially correlated, and so when a forecast is made, the future residuals are set to zero. In the model the error terms are random shocks with means zero, and so zero is used for the future values. In order for the model to predict a much worse economy, one would have to put in some large future negative shocks, like the observed shocks to the CN and CD equations in the last two quarters, which has not been done. It may be, of course, that there will be large negative shocks, due, say, to financial issues that are not in the model. The model, for example, does not account for possible credit rationing on consumers and investors from the financial distress. If there are large future negative shocks, the current baseline forecast will turn out to be too optimistic. If you have views about the size of possible shocks to some of the equations, you can put these shocks into the model and examine the results. The following are other experiments that might also be of interest.

If you think housing prices will fall further, you can decrease PSI14, which will lower PKH. This will affect consumption through the wealth variable AA (equation 89 and equations 1, 2, and 3).

Regarding the stock market, each change in the S&P 500 index of 10 points is a change in CG, the capital gains variable in the model, of about $100 billion. At the time of this writing the S&P 500 index is about 870. If you think that the index will fall, say, 100 points, you should drop the equation for CG and change CG by about -$1,000 billion at a quarterly rate (-$4,000 billion at an annual rate). See the discussion in Section 7.2 of The US Model Workbook. This will have a negative effect on real output growth because of a negative wealth effect.

A Stimulus Experiment (STIMUL with password of BASE)

The Congressional Budget Office (CBO) issued a report on March 2, 2009, which analyzed the stimulus bill ("American Recovery and Reinvestment Act of 2009," Public Law 111-5). The numbers that I have used for the present experiment are based (roughly) on the numbers in this report.

The stimulus bill has tax cuts, transfer payment increases, and increases in government purchases of goods and services. Some of the transfers are to state and local governments and some are directly to households. In the model is makes no difference whether the federal government makes transfer payments directly to households (variable TRGH) or makes them to state and local governments (variable TRGS) if the state and local governments in turn pass on the transfer payments to households (variable TRSH). To keep matters simple in the present experiment, all transfer payment increases are put into TRGH. Again, it would not matter if instead TRGS was increased and then TRSH increased by the same amount. In addition, tax cuts are taken to be increases in TRGH rather than decreases in the personal income tax rate D1G. Most of the tax cuts do not involve cutting tax rates, and so it seems better to put them in TRGH. All increases in purchases of goods and services are put in COG, federal government purchases of goods. Therefore, only two variables are changed for the stimulus experiment, TRGH and COG.

The timing of expenditures is a major issue in trying to capture the effects of any stimulus package. I have roughly followed the CBO timing for the present experiment. I have assumed that TRGH is $172 billion larger in fiscal 2009, $370 billion larger in fiscal 2010, $103 billion larger in fiscal 2011, $12 billion larger in fiscal 2012, and $11 billion larger (at an annual rate) in 2012:4. I have roughly spread these increases evenly within the four quarters of the fiscal year. For nominal government spending on goods (PG*COG) I have assumed it to be $21 billion larger at an annual rate in 2009:2, $29 billion larger at an annual rate in 2009:3, $29 billion larger in fiscal 2010, $31 billion larger is fiscal 2011, $24 billion larger in fiscal 2012, and $17 billion larger at an annual rate in 2012:4. No changes in TRGH and COG were made in 2009:1. To get the increases for COG, which is in real terms, I have divided the above increases by predicted values of PG from the baseline forecast. The total nominal increase over the four-year period of the forecast is $762 billion, of which $660 billion is in transfer payments and $102 billion is in purchases of goods.

The data set that contains this experiment is called STIMUL with a password of BASE, and you can examine this data set on the site. You can do this as follows. First, click "Solve" after "The MCD Model." Second, name your data set and create a password (say, STIMULA with a password of STIM). Third, use option 2 ("Copy another existing dataset") and enter dataset STIMUL with a password of BASE. Fourth, proceed and select option 12 ("Examine the results without solving the model"). Fifth, on the output page select MCDBASE for the comparison dataset. Then you can list and display variables for comparison. The difference between the predictions in your dataset (say STIMULA) and in MCDBASE (sometimes called MCD2BASE) are the estimated effects of the stimulus. At the end of this memo is a list of some of the results for this experiment.

It is important to note that one can have more confidince in the differences in the predictions between STIMUL and MCDBASE than in, say, the predictions in STIMUL. It could be, for example, as discussed above, that the baseline forecast is too optimistic---that some of the residuals that have been set to zero will in fact turn out to be negative and large in absolute value. However, this kind of error affects both the predictions in MCDBASE and those in STIMUL, so they cancel out when looking at differences. Put another way, estimated standard errors of multipliers are usually much smaller than estimated standard errors of forecasts.

The output below presents some of the main variables in the US model. Presented first are TRGH and COG to see the exact changes that were made. These two variables are, of course, exogenous. Presented next is real GDP and its percentage change (GDPR and PCGDPR). The peak difference in GDPR is in 2010:3, $103 billion or 3.5 percent of the baseline value. (All flow variables are at quarterly rates.) Between 2009:2 and 2010:1 the growth rates (PCGDPR) are between 2.3 and 3.9 percentage points larger. (All growth rates are at annual rates.) Presented next are the unemployment rate and the jobs variable (UR and JF). The peak difference is in 2010:3, where the unemployment rate is 1.8 percentage points lower. The peak difference in jobs is in 2010:4 at 3.85 million jobs.

Presented next are the GDP deflator and its percentage change (GDPD and PCGDPD). The largest difference in the inflation rate is in 2010:4, where PCGDPD is 0.95 percentage points higher. Presented next are the federal government deficit (SGP) and the federal government debt (AG). The deficit difference peaks at $87 billion in 2009:3. The debt by the end of 2012 is $566 billion larger. This increase is less than the $762 billion stimulus increase because of the increased tax collections. Offsetting this somewhat is that fact that interest payments of the federal government are larger. The short term interest rate (RS) and federal government interest payments (INTG) are present next (and last). Interest payments are $8 billion larger by the end of 2012. The short term interest rate, RS, has a peak difference of 0.9 percentage points in 2010:4. The Fed is predicted to raise interest rates somewhat in the more expansive economy.

An interesting feature of the results is that in 2011 and 2012 real GDP growth rates are larger in the baseline case than in the stimulus case. As the stimulus measures wear down, the growth of the economy is negatively affected. There are also in the stimulus case in 2011 and 2012 negative stock effects (durable stock, housing stock, and capital stock), negative effects from the higher price level, and negative effects from higher interest rates, which are the result of the more expansionary economy in 2009 and 2010. By the end of 2012 the number of jobs (JF) is slightly lower in the stimulus case than in the baseline case.

Other Stimulus Experiments to Run

It is easy to run alternative stimulus experiments on the site. The simplest thing to do, as discussed above, is to put all the changes in TRGH and COG. Remember that TRGH is in nominal terms and COG is in real terms.

There are also two other changes that might be of interest to make. One is to raise tax rates in 2011 and 2012, say the federal personal income tax rate D1G. There is current discussion that some taxes will have to be raised in 2011 and 2012 to keep the federal government deficit under control

Another change is to try to account for the bailout bill. If, say, the various bailouts result in a loss of $200 billion to the federal government, this is probably best accounted for by changing exogenous variable TRFG in the model. TRFG is the level of transfer payments from firms to the federal government. In 2008:4 this level was $9.35 billion at an quarterly rate, and it has been assumed to remain at this level throughout the forecast period. If there is a $200 billion loss, say spread evenly throughout 2010, then TRFG for each quarter of 2010 should be changed to -40.65, which is 9.35 less 50.0. The federal government loss is essentially a negative tax to corporations, which can be accounted for by changing TRFG. Decreasing TRFG increases cororate profits, which increases dividends, which increases household disposable income. This effect is, however, quite modest in the model because dividends respond slowly to profit changes. If you run this experiment you will see that it has a modest effect on real GDP. It mostly just increases the federal government deficit in 2010 (variable SGP) and the federal government debt (variable AG) from 2010 on. Federal interest payments are larger from 2010 on because of the larger federal debt.