A Simulated Solution to Bad Budget Forecasts

The upcoming debate over the “fiscal cliff” will once again shed the spotlight on budget forecasting. The Congressional Budget Office and other forecasters will be asked to predict revenues, expenditures and deficits under various policy scenarios. Unfortunately, these projections have developed a bad reputation, despite the tremendous effort expended in producing them.

The most infamous case of budget forecasting error occurred in 2001. At the time, CBO anticipated $5.6 trillion in surpluses over the next 10 years. The projection was used to justify $1.6 trillion in tax cuts. While those reductions were supposed to leave us with $4 trillion in surpluses over the Fiscal 2002-2011 period, the actual result was a cumulative deficit of $6.1 trillion.

Does this mean that CBO staffers are incompetent? Hardly. CBO is staffed by highly professional economists and budget analysts who strive to do thorough, objective work. Unfortunately, they can’t do the impossible, which is to accurately forecast economic growth, inflation, interest rates and policy changes over a 10- year window. If they could do that, they would probably be playing the market instead of drawing government salaries.

Since we can’t predict the macroeconomic future, what are our options? Abandoning forecasting is not one: we need some kind of projections to assess various policy alternatives as well as the risks inherent in maintaining the status quo. So we are not here to kill budget forecasting; instead we are here to save it.

And the way to save it is to replace point estimates of macroeconomic variables with distributions. While we cannot know next year’s rate of economic growth with precision, we can be very confident that it will be within a range of -3 percent to +6 percent. We can also have high confidence that interest and inflation rates will be within discrete bounds.

Instead of a single budget forecast, we propose that CBO generate a budget forecast distribution using a large number of randomly selected macroeconomic values chosen from within reasonable ranges. The results will give policymakers and the public a full range of plausible outcomes.

A common but unwarranted criticism of simulation is that the results are only as accurate as the random scenarios upon which they are based.  That is wrong. Consider the simple act of shaking a ladder before climbing on it. You are in effect running a simulation of the ladder’s response to bombardment by a distribution of random forces. Unfortunately the distribution of forces when you shake a ladder is far different than when you climb on one. So are you going to stop shaking ladders now that you know you have been using the wrong distribution for all these years? We would argue that not only is ladder-shaking a good thing to do, but, as in the current environment, when our country’s ladder is at the edge of a cliff, shaking should be mandatory.

While CBO is generally expected to provide a current law forecast, it also releases intermediate and long-term forecasts based on widely expected policy changes. A budget simulation should also incorporate various likely policy alternatives, with probabilities assigned to each one. These probabilities could be derived from political markets like InTrade. If handicapping policy changes is a bridge too far for CBO, it could release sufficient details of the budget simulation to allow private analysts to run them with their own probability estimates.

While this may all seem complicated and time consuming, recent improvements in processor speeds and software make it easy. A number of software firms offer high performance, easy to use simulation tools – many of which interface with popular electronic spreadsheets. At ProbabilityManagement.org, we are developing standardized distributions that can be represented in a compact format – making it easy for analysts to share sets of assumptions with each other and the general public. Meanwhile, the recently released Public Sector Credit Framework (PSCF) offers a free, open source approach to generating budget distributions.

Since budget simulations yield a distribution of future debt-to-GDP and other fiscal ratios, they can also be used by rating agencies and investors to assess the government’s fiscal health. If certain fiscal ratio levels are associated with default, budget distributions can be used to forecast the probability of a fiscal crisis. For example, if a rating agency were to conclude that the US cannot sustain a debt-to-GDP ratio greater than 180 percent, analysts could simply review the distribution of model outcomes to measure the probability of reaching this unsustainable level, at every maturity. By relying on budget distributions, rating agencies may be able to avoid the kind of controversy that accompanied/plagued S&P’s downgrade of US Treasuries in 2011.

Simulation techniques have revolutionized many industries in recent decades. Supply chain management, oil exploration, and financial portfolio management have all benefited by moving away from flawed thinking based on averages to an enlightened understanding of risks and opportunities provided by distributions. Now it is time for the systemically important task of budget forecasting to benefit from this 21st Century technology.