Even though recessions are becoming less frequent (e.g., great moderation), state revenue estimates are increasing inaccurate. I have covered Iowa revenue estimates before and it turns out the same pattern is being exhibited nationwide:

In 2009, the median error among states was an overestimate of 10.2 percent. That equated to an unexpected revenue shortfall of nearly $50 billion in person income, corporate income and sales tax revenues. In a year when state policy makers faced $63 billion in midyear budget gaps–coming atop $47 billion they had already closed when crafting their budgets–this misestimate posed a significant challenge.

The similar pattern between Iowa and the nation is remarkable. Here is a recent graph that shows the relationship between Iowa revenue estimates and actual revenues for December:

State net revenue in millions

Project and actual are the same curves, but project is a bit lagged. A similar graph for all states shows a similar pattern:

Pew Center on the States/Nelson A. Rockefeller Institute for Government, Cracks in the Crystal Ball, Exhibit 4

Again, revenues are overestimated at the beginning of a recession, underestimated during recoveries.

Pew also suggests these estimates are becoming more inaccurate:

During the 1990-92 revenue crisis, 25 percent of all state forecasts fell short by 5 percent or more. During the 2001-03 downturn, 45 percent of all state forecasts were off by 5 percent or more. In 2009, 70 percent of all forecasts overestimated revenues by 5 percent or more

Much of the problem stems from a basic statistical issue, it’s rare that the economy turns. Since 1991, the economy has only made six “turns”–we’ve come out of three recessions and recovered from three.* Even very sophisticated statistical models will likely miss these “turns”. So what can we do to ameliorate these difficulties?

The report highlights four promising approaches:

1. Analyzing Errors and refining assumptions
2. Making frequent estimates
3. Ensuring independence from the political process
4. Adding expertise

Each are a good suggestion. I really can’t fathom providing accurate estimates without #3 and #4 being met.

A fifth and sixth recommendation should be:

• Incorporating estimated errors
• Improving recent economic data

My fifth recommendation is somewhat technical, but important. To illustrate the importance, consider projecting a path of a hurricane. Below is an example of such projections. The projections show the hurricane will hit New Orleans (this is 2008′s Hurricane Gustav). But notice the margin of error around the projections? These are the estimated errors, and they are very important to forecasting.

Projected path of Hurricane Gustav, via watercrunch.com

There are equally important to budget forecasts. If state revenue estimates are truly becoming more inaccurate, then that means these errors are becoming more important. If revenue projections aren’t an exact science, then perhaps we should better present the data as an inexact science. Gustav? If you recall, didn’t land in New Orleans, but in nearby Cocodrie–within the range of the hurricane’s predicted path.

My sixth recommendation is also important. There are tremendous resources at the federal level that permit fairly accurate projections. As economies change, we may need to look at different sets of data to forecast future economic fluctuations. Some recent innovative practices is using road tolls to “nowcast” the business cycle.

As expected, the Iowa Revenue Estimating Conference are estimating higher state tax revenues for the current and next year than previously thought.

Last year, I predicted these revenue estimates would be underestimated. These models tend not to predict changes in the economy, so they tend to overestimate revenues during recessions and underestimate revenues during economic expansion.

Above is an updated graph comparing net tax revenues with the December projections. Notice that project tax revenues almost appear to be the exact copy of actual revenues–just shifted to the right. Projected revenues seem to be a better lagging indicator than predictor.

State net revenue in millions

The new estimates anticipate fiscal year 2011 net revenue will be at nominal levels close to fiscal year 2007. However, REC’s estimates tend not to “pivot” with the economy. As the economic data indicates the recession is ending, we should expect REC to begin underestimating net revenues.

It’ll be August 2011 before we know if the estimates are correct. If anything, I suspect it’s an underestimate. The bigger story was current year tax estimates came in high enough so further cuts aren’t needed. Although, we could have avoided some of the mess through better forecasts.

It’s a big deal. If REC says revenues will be high, the legislature will probably spend a lot. It hurts when the REC, say, overestimates. We’re then left with a deficit we have to implement drastic cuts.

Below is a graph plotting nominal actual versus nominal projected net revenue between 2003 and 2009. The projections were taken from REC’s December* meeting the prior year. The mean absolute percent error between forecast and actual net revenue was a modest 3.9%.

State net revenue in millions

The current state budget is slightly over $6 billion, so each percent error is roughly $60 million. REC’s current estimates of $5.4 billion in revenues will likely error by $210.6 million.

Tax receipt forecasts were slightly more accurate, ff by an average of 3.7%. That means the projections fall apart on other receipts (e.g., judicial revenue) or fund transfers (e.g., lottery payments). Those revenues are probably more cyclical (e.g., liquor tax revenues increase when the economy declines) that’s harder to predict.

Tax receipts in millions

Nevertheless, REC’s margin is quite large. The problem seems to be two-fold. First, REC underestimated revenues between 2004 and 2008 which is understandable since they use conservative estimates.

Second, REC doesn’t pivot with the economy. As Iowa’s economy began to sank REC went from underestimating revenues to overestimating it. A similar pattern is apparent from 2003 to 2004 as Iowa’s economy recovered. REC, who overestimated revenues in 2003, underestimated them  through 2008.

The Legislative Services Agency (LSA) has a done a study before on the accuracy of REC in the late 80s/early 90s. The data seems to show show that REC overestimated revenues in 1988, but the methodology isn’t clear enough for me to be sure.

*Prior to 2002 they had a different meeting schedule. I used the February projections since it was closest to the December meeting.

This post was updated after an error was discovered.