Breaking Down the Federal Budget: New Tools from RAND

Deanna Lee

You're listening to Policy Minded, a podcast by RAND. I'm Deanna Lee. The federal government collects and spends trillions of dollars every year, making policy choices that affect every household, every business, and every community in America. But understanding exactly how those decisions shape our lives, well, it's complicated. That's where RAND comes in.

Our researchers have developed a new suite of tools to help policymakers and all Americans understand just how much can change with every line in the federal budget. This effort, the RAND von Furstenberg Family Budget Model Initiative, is co-directed by senior economist Jeffrey Wenger, who joins me now to discuss these groundbreaking new tools. Jeff, welcome.

Jeffrey Wenger

Well, thank you for having me. I appreciate it.

Deanna Lee

Okay, before we dive into the RAND Budget Model, Jeff, tell me a little bit about your background, what brought you to RAND, how you got here.

Jeffrey Wenger

Well, I'm an economist. I went to University of North Carolina, Chapel Hill, a long time ago. And then I went into Washington, D.C., and worked at a think tank in Washington D.C., and then moved from there to become an academic. And I spent more than 10 years as an assistant and associate professor and wanted to do more policy-relevant work, wanted to things that were bigger in scope, more complicated, more challenging, really wanted to choose. Jump in deeply into policy analysis in a way that working by yourself in your academic office isn't going to allow you. So I came to RAND and did that about 12 years ago and been working on a variety of complex, difficult policy questions ever since.

Deanna Lee

Talking about things that are large in scope, I think the RAND Budget Model fits that bill. So let's get into it. What is the RAND Budget Model?

Jeffrey Wenger

The RAND Budget Model is a suite of tools, but that doesn't really help people understand what the RAND Budget Model really is. What people really want to know is how policy is going to affect their lives. And individually, they're really good at figuring that out. We can look at many, many different policy examples ranging from food stamp benefits to health care subsidies to child tax credits, all of which will impact the individual and every individual is likely to be able to know. How that's going to affect them individually.

What they don't know is how that is going to effect the marketplace writ large. So what the Budget Model really does is it takes the individual level responses, the individual level changes, and, through a variety of different tools, aggregates them up to market level analysis so that we can understand what's going happen in the market writ large, so that's one dimension. And the second dimension is that we want to forecast what's going to happen over the long haul. We want know 5, 10, 15, 20 years from now what's going to play out, how that's going to impact individuals, households, corporations, states, legislators. All of these things are part of the remit of the Budget Model.

Deanna Lee

And what makes the Budget Model different from models that already exist or from other approaches that have already existed to tackle some of these same issues or questions?

Jeffrey Wenger

The way I like to think about the Budget Model at RAND, as compared to other budget models, is we are like the DNA of the budget. We drill down very, very deeply into the specific lines of code, whether that's the tax code or how the spending gets done. We have many tools as part of the Budget Model, but the most important tools are the tax module and the spending module. That's what we call them at RAND.

The tax module literally takes the entire U.S. tax code, deconstructs it line by line using AI tools, and then provides an analytic framework so that we can understand how every single one of those lines of the tax code links up to the IRS tax forms and links up to individuals and corporations. So it's molecular in its structure. It's not … It's not as macro as some of these other models where, you know, you might take a standard budget model and there might be three or four different parameters, different elements of the modeling that get adjusted. Well, we can adjust corporate tax rates from here to here, and then we can see what happens. What we're going to do is we're gonna say, okay, well, what happens if this particular corporate tax rate gets reduced? How will that change? How does that interact with this other corporate tax change? And we're going to be able to do that on a deep, molecular level.

Deanna Lee

And so, given that level of granularity, for lack of a better word, who was this tool designed for? What kind of questions did you have? You've given us an example already, but what kind of questions did you have in mind when it was being built, and who did you have in mind?

Jeffrey Wenger

It's a really good question because we really built the tool—it was one of those situations where all of a sudden we now had the capacity to investigate the tax code in a way that we've never been able to do it before because of these AI tools. It was almost as if we discovered the tool and then we looked for an application. Most of the time at RAND we have a problem and then figure out tools that we need to solve the problem. This was sort of the reverse. It was sort of … Carter Price, who's my co-lead on the Budget Model, he's a mathematician and he's just got this way of thinking about the granularity of policy analysis and the granular of the tax code and his capacity to use these AI tools that he in essence built the tool and then he said, okay, what should we do with this? And then all of a sudden we had a thousand flowers blooming because we could do myriad things. … We're going back in time to 1925, where we're gonna take the tax code, going back and basically since we've had the Internal Revenue Service, since we had individual taxation, and looking at the history of it, looking how it's changed.

There's so many things that you can do once you've built this tool, and so we're just getting started in fully investigating the power of what we've got. And we're still building it out. We're ever expanding it. We're linking it to tax forms. We're then linking the tax forms to corporate data so we can understand what happens to corporations. We're linking that to individual tax filing. It just keeps growing and growing and growing.

Deanna Lee

Okay, that's very interesting. And I do want to ask you a little bit more about these AI tools behind the model. So what else can you tell us, bearing in mind that I'm not an economist, I'm not a technologist, and maybe a lot of our listeners aren't either.

Jeffrey Wenger

Yeah, so maybe some of our listeners have used the large language models, this sort of chat GPTs or Claude, they know that you can upload a document and you can have it summarize that document. This is not that; this is actually more technically detailed. In essence, what you have to do is you have to train or build an AI that can parse the language, literally take the the tax code, which is Title 26, break it down into small chunks, train the AI on it, and then begin coding the elements of the tax code. This is an entity, it's also, you know, a firm is an entity. What kind of firm? Well, it is an S corporation. Then you train all the AI and all the things that are S corporations, and then how do S corporations get linked to different outcomes in the tax. So you're literally building this entity relationship policy space. And then once you've got that space, you can now start doing investigations on it.

So you've broken down the tax code using these AI tools, you've built out all the relationships between the entities and the policies, and now you can start investigating them and linking them to all the other portions of the tax codes. So in essence, what we can see is tax complexity directly, but we can see the interconnections of specific components of the tax code. That's never been done before. And that allows us to do really interesting work around, oh, here's this orphan tax policy, why does it exist? Who files for it? How did it get into the tax code to begin with? When did it arrive? When does it expire? Granular detailed item-by-item analysis like that. And then we can aggregate that back up and you get a larger picture of what's going on in the U.S. economy.

Deanna Lee

Okay. And I imagine that you're constantly—or perhaps constantly isn't the right word—but consistently training and kind of tweaking the AI behind it, correct?

Jeffrey Wenger

Absolutely. We have to update the code because the [tax] code gets updated. Every time Congress passes a bill, something changes in the tax code. So you go from the One Big Beautiful Bill Act, and now we have a new major revisions to the tax code. All of those need to be incorporated and analyzed and figured out. And it doesn't just end there, right? Because of course we're doing work related to complexity of the tax code. A lot of the complexity of the tax code is due to what we call vague language. The tax code says a "significant increase" in expenditure. That's the language in the bill when Congress passes it. But someone now needs to interpret what "significant" means. And that's part of the regulatory process. And so one of the things that we're looking to do in the near future is start linking this to the regulations. Because now you can take the tax code. That's got all this vague language in it, link it to the regulations, and then see how the regulatory authorities actually implemented the tax code. So, it's just this hydra. It just keeps growing and growing and growing and growing in complexity and … growing in analytic power.

Deanna Lee

Now that we have a little bit of a better sense about what the RAND Budget Model is, let's talk about one of the questions you tackled using it, and that's how to reduce the federal debt burden. The federal debt has topped $38 trillion, and the U.S. now spends more than $1 trillion annually on interest payments alone. What does this problem mean for the country and for Americans?

Jeffrey Wenger

The United States is in a really unique position because we are the reserve currency of the world that we get to run deficits that other countries would be hampered by. And the question is, at what level of deficit do we run before the bond markets and the people who are willing to loan us that money lose faith in the United States? So we have a very unique and a very difficult question to answer with respect to that, right? At what point—you said we're at $38 trillion worth of debt … that debt to GDP ratio, so how much the total economy produces in a single year is about $38 trillion. So our debt is now equal to one year's worth of total output from our economy. That's historically really high for us. Is it catastrophic? Probably not. Not yet. Are we in the danger zone? Are we the yellow zone? Warning signs are flashing, yes, I think that that's really likely. And we're seeing that by some of the bond rates, the amount of money, the amount of interest that we have to pay people to get them to take new debt from the United States, those interest rates are going up. Why are those interest rates going up? Because they have less confidence in the United States to pay back their debts going forward.

I won't get into federal policy right now. There's a huge call for the Federal Reserve Bank to lower interest rates, but they don't control long-term interest rates. They just control short-term interests rates. Long-term-interest rates are set by the market, and the market right now is concerned that the U.S. Is going to have a difficult time paying back its debt. So what do you do about it? You can raise taxes and pay down your debt, right? That's one of the things you can do, or you can reduce spending. It's really mechanical, right. It's just, you pay down the debt by either raising taxes, by spending less, or by becoming more productive. If we do that, then we can sort of, you know, grow our way out of the deficit. The tax tools that I talked about, this is just weren't used in this portion of the analysis, right? So we didn't give them a suite of policy recommendations that they could implement on the tax side. What we really said was, historically, this is where we were in the past. In the immediate aftermath of the Second World War, we had a debt to GDP ratio that was about 100 percent where we are today. And we paid it off and we did it in a variety of ways. We did all three. Productivity growth increased, our expenditures went down, our taxes went up, and we were able to pay that debt-to-GDP ratio down.

So the point of that previous work was, it's been done in the past, it's not so scary, we just have to have the political will to do it. And there are some different tools that we can implement to make it more effective. In the short run, I think it's gonna be very difficult to cut expenditures. You have too many people who are reliant on Medicare and Medicaid, and social security that are already retired, that you can't really cut their benefits because it will just result in direct hardship for them. So I think the revenue side, we need to raise taxes and we need a figure out ways to increase productivity. Hopefully AI does some of that for us. That way we can again grow our way out of this debt to GDP problem.

Deanna Lee

And in thinking about the challenge we're facing today compared to post-World War II, of which you mentioned, did the Budget Model provide any revelations suggested that the solutions today are different than they were after World War II?

Jeffrey Wenger

Yeah, I think that that's, you know, we pointed to a number of things that are fundamentally different in the post-immediate aftermath of the second World War. We had a younger population. We had growing—we had a baby boom who was entering the labor force for the first time. So our working population was going to grow dramatically. That was automatically going to increase our tax revenue. It was also going to raise our productivity levels, so we were going to put all this talent to work in the labor force. Those were very different circumstances than what we face today, where we have a constant if not shrinking labor force and a population that's dependent on social benefits like social security and Medicare. So it's a tougher—we definitely have headwinds today that we didn't have in 1945—whereas in 1945 we had tailwinds that policymakers could do less, make fewer hard decisions and still be successful. Now we're gonna have to make more hard decisions about what to do.

Deanna Lee

So if our listeners visit rand.org/budgetmodel, they can find this study that we're discussing on the federal debt. But they'll also find other studies on how AI affects productivity, the tax code, which we've discussed a little bit, Americans' retirement savings. Jeff, do you want to share any insights from those other studies, maybe some surprising or particularly interesting things the Budget Model revealed from other research efforts you've worked on?

Jeffrey Wenger

Absolutely. I think one of the most interesting things that the Budget Model has done recently is analyze the effect of the Retirement Savings for Americans Act. This is a proposal that's come out of the House and Senate. It's a bipartisan proposal. The goal of the policy is to get retirement savings into people's hands—workers, who currently work in jobs where their employer does not provide them access to a retirement savings plan. So—lots of people who work and their employer doesn't give them a pension or a 401k, or other kinds of retirement savings—what this policy would do would be to make a government plan available to them. All the workers who work for those employers would have access to a government plan. The individual would make a contribution and the government would match that contribution. The interesting thing about that policy is, you know, how many people are going be affected, and it's millions. That's interesting in and of itself, but that's relatively straightforward. I think the unique insight that the Budget Model was able to provide was looking out 15, 20, 25 years into the future: what happens when you've had 25 years worth of retirement accumulations to your net worth and to your dependency on other types of social programs.

So let me step back for just a second. The Congressional Budget Office is assigned the task of scoring legislation that comes out of Congress. So when a proposed piece of legislation comes out of congress, it goes to the Congressional Budget Office, and the CBO says, this is how much this will cost. If it's revenue neutral, it doesn't cost too much money, then it gets a special set of provisions in terms of how it gets voted on in Congress. If it costs money, then you get another set of provisions about how it gets voted on in Congress. So there's these legislative rules. Typically, the CBO scores a bill, determines what the cost will be, on a 10-year window. What does this mean? It means that any bill that invests in a 5-year-old is not likely to see returns on that investment after 10 years, because now they're 15. They haven't necessarily gone into the labor force. They haven't started generating tax revenue. They don't have a job yet, et cetera, et cetera. This methodology, in essence, that the CBO adopts predominantly, excludes these kinds of long-term investments from reaping the benefits that they would reap 15, 20, 25 years down the line.

So what we did is we said, let's look at this going out many, many years. And what will happen as a result of doing that, we found, asset accumulation will be significant even for the lowest-income workers. Because we have this time value of money and a match from the federal government. In other words, you get a five or seven or eight percent rate of return on that money every single year—and you do that for many, many years—even a small amount of money will turn into $100,000. That means when you go to retire, you're not gonna be as dependent on Medicaid as you otherwise would have been. Turns out that fully pays for itself. Over that 25-year time horizon, which was really surprising to people—that government investing in workers with limited access to retirement savings today ultimately pays for itself by reducing dependency on Medicaid expenditures 25 years from now. And you wouldn't have seen that if you just did this through CBO, because the CBO I was only going to do a 10 year window of analysis.

Deanna Lee

This leads me to a broader question, which is, does this forecasting capability, which is part of the RAND Budget Model, is this a game changer for understanding kind of the long-term implications of policies?

Jeffrey Wenger

Absolutely. We have to be careful to make very broad blanket statements. It's not that the CBO never does an analysis that extends beyond 10 years. They do. They do it rarely. They don't do it for all policies, but some policies require it, but not every policy. And CBO is just bound by a certain set of rules that we're not bound by. And so we can make different decisions, and we can make different forecasts, and we can make alternate projections. You know, one of the things that CBO has to do is they have to take current law into account. If the bill that's up for a vote says, we're going to repeal all of these things after five years and the revenue is going to go back up, that's the law as written. CBO must take that into consideration, even though they know that there's virtually no likelihood that that's going to get repealed, and taxes are going to go up. You know? What choice do they have? As a RAND budget modeler, I don't have to make that assumption. I can provide two sets of analyses—one that says if you repeal, this was what will happen and that should correspond to what CBO says, plus our differences in modeling. But I can also say, what happens if we don't repeal this after five years, what will happen to this? And you can assess the probability of which one of those two things is more likely, repeal or non-repeal.

Deanna Lee

In building these models, I imagine it needs to be designed or trained to deal with shocks or surprises in addition to kind of inputting the large knowledge base we have from the past—from the tax code, for example. COVID-19 might be a good example of one of the shocks that happened that you couldn't account for. So how do you deal with this kind of tension of having existing content and having a lot of unknowns.

Jeffrey Wenger

Whenever you're doing modeling or forecasting, it's always much easier to predict something that's similar to something that's happened in the past. Because you've got an example. If we change the corporate tax rate from 22 percent to 25 percent, this is, in general, how corporations will change their behavior. They'll make these types of investments. They'll have an incentive to offshore income, to offshore intellectual property, whatever we think will be the impact. But we've seen changes in corporate tax rates many times over many years, and we've observed how people have changed their behavior. When we have a pandemic, now we're really in a completely different space. So we have to start thinking about new ways of modeling that change in behavior.

So I'll give you a really interesting example. When COVID hit, it didn't dawn on a lot of people until later that our consumption behavior was gonna completely shift. Because COVID was a socially spread pandemic, all social interactions fundamentally changed, which meant that anything that you consumed that required a social interaction, a restaurant meal, a massage, a haircut, whatever it was that required a social reaction was suddenly going to shift. And we were gonna take all of those monies that we were spending on those things and buy goods because those were not socially. That generated a huge response that was unprecedented, but at least could be theorized about given the nature of the pandemic itself.

So what happens in those situations is you have to rely on theory and you have to rely on the specific circumstances of the shock to determine what's likely to be happening in the long run. That, of course, put pressure on supply chains, because now we needed to ship this many more goods. That, of course, put pressure on inflation because we had to produce that many more goods that we didn't necessarily have capacity to produce. And you could then have told yourself a story once you understood the logic of what was likely to happen. So you move from empirical example to theoretical example or theoretical impact, which is really the only tool we have under those circumstances.

Deanna Lee

OK, let's talk about what's next for the Budget Model. What are you working on right now that you want to tell us about?

Jeffrey Wenger

One of the things we're working on right now … we're doing two, I think, very interesting projects. The first of which is corporate taxation. So I talked about how we've got a tax tool that allows us to go very molecular into the tax code. We're tracking that tax code back to 1925, all the way through to 2024, where we're at today—2025 for the legislation, 2024 for the IRS data. The interesting thing is what's happened to corporate taxes over this time period. I don't want to give too much away because we're still doing the work. So we're tracking what happened to corporate taxes and why.

Corporate taxes have gone down a lot. They peaked in about the 1950. I don't think that's anything hidden. But why? What were the provisions? What was the impetus for that change? How did it occur? When did it occur? Has the tax code become increasingly complex? Has it gotten simpler? When were the major periods of time when the tax code changed? And then what we really want to link it to is some causal understanding of what were the driving factors that did this. I think it's going to be a really interesting piece. We're going to try and link up a variety of different data sets to figure out what was driving the reduction in corporate taxation over the last 80 years in the United States, which is a paper that wouldn't be possible without these fine-grained, analytic tools developed by AI.

Deanna Lee

Great. And what was the second project you were working on?

Jeffrey Wenger

The second project we're working on is looking at R&D expenditure in the United States. One of the ways in which you can get out of the debt-to-GDP problem that we've got is to accelerate productivity growth. One of ways in we know productivity gets enhanced is by developing new technologies, developing new techniques of producing things. Technological investment has a big bang for its buck. One of the things we're doing with the the AI is tracking down all of the authorities inside the federal government that are authorized to spend on R&D. We know that, for example, National Institutes of Health spend a lot of money on research and development in the health space. National Science Foundation spends lots of money in science and technology. What are all these authorities? What do they look like? How many people are allowed? What's going on? But then tracking what's actually happened as changes in R&D spending have occurred in the United States, and pushing that down all the way to the local level. So we can take that money that we're spending on R&D and figure out where we're spending it and what impact it's had on local economies and what the spillover effects are going to be, what the investment returns are going be all the down to the local level. If you want to grow the economy, what's the proper mix of R&D? Where should we be spending our R&D dollars? Should we be sending it on basic science? Should we spending it on converting existing technologies into application? Those are all different kinds of things that we can do policy-wise.

Deanna Lee

Wow. Okay. And I imagine nothing like that's ever been done before, at least that sort of level of detail.

Jeffrey Wenger

Not to my knowledge, no.

Deanna Lee

I think you've really given us some insight into the possibilities of the RAND Budget Model. It seems endless. How do you see this being used in the future?

Jeffrey Wenger

Ideally, you know, we'll continue to build the complexity of the model in the background. We need to have a user interface that allows people to more readily utilize the tool. So something very, very complicated in the background can have a relatively simple user interface that says, I wanna know … where the electric vehicle tax credit went. And they type in "electric vehicle tax credit," and it shows you a map of that over time. And you can get very simple answers out of a very complex set of tools. What we'd ultimately like to do is build a user interface that takes all of this complexity in the background and allows government entities, Congressional Budget Office, Joint Committee on Taxation, legislative staff, to be able to use those tools to better inform how to make policy, look at changes historically, determine what the likely impacts of that change would be. Those are the kinds of things that make it really user-friendly and make it much more accessible to the policymakers and the people that we would like to see use it. And that's just the federal level. Of course, there are many, many state governments who lack any capacity to do similar things. They don't have the legislative staff that our national government has. They don t have the depth of experience at the state level often. And so taking that down to the state-level ultimately would be, you know, is really long-term dream because the state and the federal taxes interact in complex ways, you have to model that, you have model each of the states' tax codes, it goes on and on. But that's really the dream in the long run is to be able to take a fully integrated system, state, federal tax, and model what would be happening if you were to change something at the federal level all the way down to the state level.

Deanna Lee

That's a really exciting prospect to think about, to be sure. You've highlighted a lot of the ways that the RAND Budget Model can be applied to problems that affect all of us, but if you had to sum it up for listeners, why should they care about this?

Jeffrey Wenger

I think it goes back to what we sort of opened this discussion on is I think individuals, individual people, households are very good at figuring out how a policy impacted their budget, their household budget. What's really, really difficult sometimes is to figure out how it aggregates up to the marketplace. How is it going to affect your local access to health care? How is this going to affect your local food bank or your access to healthy food. It's not just about the individuals, but it's the aggregation of the individual impacts to the market level that the Budget Model can really give us insights to and tell us something about what's going to happen—not just in your local community, but potentially in your state or across the country as a result of it. That's something I don't think a lot of people spend a ton of time thinking about. That's our job. But that's what this tool can really bring to folks, is what's going to happen in the aggregate to the economy writ large.

Deanna Lee

And am I right to assume that a lot of those questions now remain unanswered until after a policy has been passed and kind of been out in the wild for, you know, five or ten years?

Jeffrey Wenger

That will always be the case, right? Because we can make forecasts and our forecasts will, like all forecasts, will be wrong. All forecasts are wrong, they're just a matter of degree. But at the end of the day, we'd really like to be able to take historical changes and use those to make forecasts about what would be likely to happen based on what's happened in the past. And so we can get better and better forecasts as we get better data. And as we get these data linkages tighter and tighter, we'll be able to understand the actual mechanisms that underlie these changes and what's likely to happen. Most of the time, we get it fairly accurate. I mean, we're wrong, but we're only wrong by a small amount. And so we can anticipate what's likely to happen as a result of these policy changes. And I think we'll better at doing that when we have these more detailed tools and these better data and better modeling techniques.

Deanna Lee

Absolutely, and it sounds like there's much more to come. So if someone's interested in the RAND budget model, how can they learn more or maybe even get in touch?

Jeffrey Wenger

Sure, you can go to rand.org/budgetmodel, so it's all one word, no capitalization. Or, you can contact me directly at my email, which is listed on the RAND website.

Deanna Lee

As Jeff mentioned, rand.org/budgetmodel will give you a lot more information about not only the Budget Model itself, but the research we discussed today. You can also visit rand.org/policyminded to find a transcript of today's episode and links to the reports that we discussed. Thank you to everyone listening. Jeff, thank you so much for joining us today.

Jeffrey Wenger

My pleasure to be here.

Deanna Lee

Today's episode was produced by me, Deanna Lee. I also recorded the episode along with Emily Ashenfelter, who is our editor. Pete Wilmoth is the director of Digital Outreach at RAND.

Jeffrey Wenger

RAND is a nonprofit institution that helps improve policy and decisionmaking through research and analysis.