This page is the first installment of a macroeconomic simulator. It lets you pick a shock—which could be policy like the Fed raises rates or events like a fall in the stock market—and simulate the impact on a wide range of macroeconomic and financial variables. For now it is based on FRB/US, the Federal Reserve Board’s workhorse model of the U.S. economy. It traces out the consequences of different shocks as a difference from the baseline—so it tells you, say, that something raises GDP by 0.5 percent but not what GDP would actually be. The defaults for all of this are set in a reasonable manner but many of them can be changed if you choose “Customize” and alter how the Fed responds, expectations are formed, how long the shock lasts, and how it is paid for. You can also add more shocks at the same time (for simplicity it just adds them together—which is approximately right) and compare different scenarios.
In the future this page will add more macroeconomic and possibly other models as well. Other related tools are the Fiscal Explorer and the Taylor Rule simulator.
Loading the model’s responses…
Every other variable the model reports, on the same scenario. Click a small chart to enlarge it and download it.
FRB/US is the large-scale model the Federal Reserve Board’s staff has used for forecasting and policy analysis since 1996: 284 equations, about 60 of them estimated from decades of data, describing households, firms, government, financial markets and the rest of the world, with an explicit treatment of how expectations are formed. The Board publishes the model, its equations and a dataset. This page runs that package through the Board’s own Python platform with four changes: the policy-rate equation is swapped for the rule you choose; for the tax shock, the personal tax-rate equation is rewritten so the shock can shift it; for the tariff shock, the equation for the price of business investment is rewritten so that price stays on its own path and moves only with the duty on imported equipment (see that shock below); and for the business-investment shock, investment and imports are each split into the model’s own equation plus an added piece, with the equations for GDP and the capital stock adjusted to match (see that shock below). Results are the model’s; they are not the Board’s view, and the Board’s data package carries its own disclaimer that its extrapolation is not a forecast.
Every impulse response starts from the Board’s long-run extrapolation in the data package, at a point where the economy is in a steady state: output at potential, unemployment at its natural rate, inflation at the 2 percent target, the funds rate at the neutral real rate plus the target. Tracking residuals make the model reproduce that baseline exactly. Then the shock is applied, and each chart shows the shocked path minus the baseline: percentage points for rates and inflation, percent for real quantities and asset prices, percent of GDP for fiscal variables. The size slider scales the stored runs. The model is close to linear for moderate shocks, so a 0.5× shock is half the 1× run, and sizes between the stored points (1×, 2× and their negatives, and 3× for stock prices) are interpolated; any such scaling is noted under the headline. Where the model did not solve one stored size, the page works from a size that did solve and says so. Where no run in that direction solved, it says the combination did not solve rather than reversing a run in the other direction: the model is not symmetric, most of all at the lower bound, where the Fed can raise rates but not cut them. At the lower bound the page does no scaling at all: the floor on the funds rate makes the model nonlinear, so the slider stops only at the stored sizes, every chart is a run solved at exactly that size, and a run that did not solve is shown as not solved.
Each shock lasts one year or is permanent; “Exactly how long” under Customize also offers two and four years, and for the Fed’s own move one quarter, two years and three years instead. A temporary shock simply stops when its period ends, and the shaded band on the charts marks that period. A permanent shock stays in force for the whole solution period, 50 years for the forward-looking runs, so nothing in the ten years shown is the shock wearing off.
Each tile names a variable, and the slider sets how far it moves and in which direction. One unit of each, described here as a rise (a negative size is the same move downward, solved as its own run, since the model’s response to a fall is not always the mirror image of its response to a rise):
Federal funds rate: the funds rate is held 1 point above its baseline path for the chosen period, then the chosen rule takes over. Permanently, it stays there whatever happens. Government purchases: real federal purchases of goods and services rise by 1 percent of real GDP. Taxes: the average personal income tax rate is shifted by enough to raise 1 percent of GDP in revenue, as a flat shift for the whole period (the slider starts at a cut of that size). Transfers to households: the trend ratio of government transfers to GDP rises by 1 point. Oil prices: the price of imported oil is $25 a barrel higher. The dollar: the real trade-weighted exchange rate is 10 percent higher. Stock prices: stock-market wealth moves 10 percent on impact, up to 30 percent either way, for one of two reasons chosen under “Why stock prices move” (explained below); the slider starts at a 10 percent fall. In the required-return version the equity premium moves, by about 0.4 of a point for a 10 percent fall; in the earnings version investors’ expected growth of real dividends moves, also by about 0.4 of a point a year. Either move is exact with backward-looking expectations, the setting each shift is chosen in. With forward-looking expectations markets also price in how interest rates and the economy will respond, so the same shift moves stocks by a different amount: in normal times anywhere from about half the size chosen to a little more than it (from about two-thirds in the earnings version), and at the lower bound, where the Fed cannot cut, a fall can be about three times as large; the note under the headline gives the move whenever it differs from the one chosen. Long-term rates: the 10-year term premium is 1 point higher; a fall is what a bond-buying program does. Labor market reallocation: the natural rate of unemployment is 1 point higher, as when workers moving between jobs and industries take longer to match (after the pandemic, or as AI or trade reshape demand); FRB/US has no industries, so this is reallocation’s footprint on the whole economy — lower potential output, and wages that respond to a smaller gap — not workers moving between sectors.
Why stock prices move. The same rise in stock prices can come from two places, and the page lets you choose. Lower required returns: investors accept a smaller reward for the risk of holding stocks (the equity premium falls), so the same expected earnings are worth more today. In FRB/US that required return also makes up half of what finance costs businesses, so cheaper finance lifts business investment as well. Higher expected earnings: investors expect company earnings to grow faster, and nothing else about finance changes. In the model that expectation enters only the stock market, so it works through household wealth alone: families who feel richer spend more, and businesses invest more only as their sales rise. For a lasting 10 percent rise, with backward-looking expectations, the balanced-approach rule and taxes fixed, real GDP in the second year is 0.18 percent above baseline in the required-return version and 0.13 percent in the earnings version, and business investment is 0.47 percent above baseline against 0.05 percent; consumer spending rises about a quarter of a percent in both. A broad repricing of risk, with investors everywhere more willing to hold stocks, is the first version; a boom in a few companies’ expected profits, the way a new technology can lift a handful of firms, is closer to the second.
Tariffs: a 10 percent tariff on imported goods other than oil. The duties, 10 percent of non-oil goods imports as they would have been without the tariff (about 1.1 percent of GDP: FRB/US’s non-oil imports include services, so the duty base is goods’ share of them, 78 percent in 2024, from the Bureau of Economic Analysis), go to the Treasury as customs revenue and do not shrink as imports fall. Who pays them: consumers pay half, through higher prices spread over three quarters (a quarter, then 45 percent, then the rest), approximating the timing Federal Reserve staff found for the 2025 tariffs (Minton, Ray and Somale, 2026); half is the share that matches published estimates of those tariffs’ effect on core inflation. Businesses that buy equipment pay half the duty on imported capital goods, taken as 29 percent of the duty base (capital goods are 29 percent of goods imports and took 28 percent of the extra duties collected in 2025–26, Census Bureau), through a higher price of investment on the same schedule. Company profits absorb the rest, so companies bear half of the duty in all. The model’s core price level is moved on that schedule, and back out on the same schedule when a temporary tariff ends; FRB/US left to itself passes a tariff into consumer prices only over years. The price of imports including the tariff holds imports down through the model’s import equation; the measured import price itself is left alone, since it is recorded before duties and raising it would count the duty a second time, as a payment to foreigners. Under the defaults, a lasting tariff lowers GDP in the first year as the Fed responds to the jump in prices; spending that shifts from imports to U.S. producers then props it up for a few years; and dearer equipment holds back investment and productive capacity, so after ten years GDP is about 0.2 percent below baseline and capacity about 0.3 to 0.5 percent below, under the balanced-approach and inertial rules and the holds. The other two rules differ: under Taylor (1993), which leans less against the output gap, GDP is about back to baseline by then, and under the first-difference rule, which keeps tightening while inflation is above target, it is about 1.6 percent below. With backward-looking expectations stock-market wealth ends about 7 percent lower, because profits absorb part of the duty. Under the Board’s debt rule the extra revenue goes back to households as lower personal taxes, which keeps spending up for longer, but capacity is just as much lower. The model has no retaliation, no supply-chain disruption and no effect of uncertainty, all of which would lower GDP further, and the dollar moves only as its own equation says.
Immigration: the working-age population is 1 percent larger, about 3 million people. For a temporary shock it stays 1 percent larger for the period chosen and then returns to its baseline path all at once, as if the same people had arrived early; permanently, the difference lasts (the slider starts at a population 1 percent smaller, less immigration). Inflation expectations: long-run expected inflation (the 10-year expectation in the Survey of Professional Forecasters, which anchors price setting in the model) is 1 point higher — expectations coming unanchored. Credit spreads: the risk premium on BBB corporate bonds is 1 point higher, a financial-stress shock that raises business borrowing costs and, in the model, the equity premium with it. Business investment: business investment is higher by 1 percent of GDP, up to 2 percent either way, and “Share of the equipment imported” sets how much of it comes from abroad, from none to all. The added investment sits on top of the model’s own investment equation, which still responds to the economy, and its imported part on top of imports; GDP counts each dollar at today’s prices, so the direct effect of a dollar of investment on GDP is exactly its domestic share; the total effect also depends on how the rest of the economy responds (a one-year, all-domestic shock of 1 percent of GDP raises real GDP 0.87 percent on impact under the balanced-approach rule). The business-investment chart includes the added investment itself, about 9 percent of business investment for a shock of 1 percent of GDP, as well as the model’s own response. The new capital is not counted as adding to productive capacity, so this is a demand shock; a payoff in productivity is the productivity tile. The page stores the all-domestic and all-imported runs and mixes them for the shares in between, which in checks against direct runs came within 0.01 percentage point. At the lower bound only none or all is offered, because the floor on the funds rate makes the model nonlinear and the two runs cannot be mixed there. Productivity works differently from the rest. Over one year it is the Board’s standard supply shock, a one-time 1 percent rise in the level of multifactor productivity that never goes away. Over longer periods it is a rise in productivity growth, 1 point a year faster for the period chosen; permanently, faster forever, so output pulls further ahead every year.
Under Customize you choose the Fed’s behavior; the default is the balanced-approach rule. All four rules set the funds rate from inflation π (core PCE inflation over the past four quarters, as in the model), the Fed’s 2 percent target and the output gap, and all but the first-difference rule also from the neutral real rate r* (1.2 percent in the Board’s long-run baseline; with forward-looking expectations it learns, moving 5 percent of the way toward the real funds rate each quarter). In every case the rate can never fall below the model’s lower bound.
For the Fed’s own shock the choice is which rule takes over after the move. A permanent move has no rule at all: the Fed keeps the rate away from baseline whatever happens, the experiment that shows why central banks follow rules. It is offered only with backward-looking expectations. Under forward-looking expectations a permanent rate peg has no unique answer, because with the rate fixed forever nothing pins down expected inflation (the textbook indeterminacy result), and the model’s answer flips sign with the financing assumption.
FRB/US has two ways of closing the federal budget, and the choice matters for every shock that changes the deficit, which in the model is all of them.
Backward-looking is the Board’s default VAR-based version, in which expectations are formed from small forecasting equations on recent data: people learn about a policy only as it shows up in the data. Forward-looking is the model-consistent version the Board’s staff commonly uses for monetary-policy simulations: expectations in financial markets and in wage and price setting are the model’s own future solution, so an announced policy path is understood the day it is announced, while household and business spending stay backward-looking. Forward-looking runs are solved over 200 quarters, 50 years, so the end of the solution lies far beyond the ten years shown: the long-term interest rates and stock prices that markets set look decades ahead, and a permanent shock needs that long to settle. Solved 75 years out instead, permanent shocks moved GDP, unemployment, inflation and the funds rate by no more than 0.05 point in every case checked. Stock prices are the exception, because in the model the stock market looks furthest ahead: for a permanent shock, stock-market wealth can differ by up to about 1.5 percent, long-term interest rates by up to about 0.1 point, and housing and business investment by a few tenths of a percent, depending on where the solution ends, most of all for shocks that never settle down, such as productivity growth that is faster forever or extra spending with taxes fixed, so that the debt keeps growing. Read those charts for a permanent shock as approximate. In these runs the rule’s neutral real rate r* also adjusts: it moves 5 percent of the way toward the actual real funds rate each quarter, the Board’s own option for its rules, so a shock that changes the neutral rate does not leave the Fed aiming at the old one. With backward-looking expectations r* stays at the Board’s 1.2 percent.
The “at the lower bound” setting puts a floor under the funds rate exactly at its baseline path: the Fed can raise rates in response to a shock but cannot cut. The floor holds for 30 years, far beyond the ten years shown, and then lifts; that matters only for forward-looking runs, whose solution runs 50 years, since a floor lasting all 50 would rule out any cut for half a century, and the model could not solve many runs that way. That is the situation of 2009–15 and 2020–21 without having to move the baseline there, and it is why a bad shock does so much more damage in that setting: nothing stabilizes it. The forward-looking model solves at the lower bound only with taxes fixed; with the Board’s debt rule on, the runs did not converge, so that combination is grayed out.
“Add another shock” under Customize puts up to three more shocks on top of the first, each with its own size and length, all under the same Fed response, expectations and financing. Each shock can be used once: for a bigger tariff, say, set a bigger size rather than adding a second one. The page adds up each shock’s own run. In normal times FRB/US is close to linear, so the sum is close to what a single run with every shock at once gives: in checks against such runs, with two or three shocks, temporary and permanent, up to twice the usual size and under both kinds of expectations, the two usually came within about 0.05 percentage point on GDP, unemployment, inflation and the funds rate for moderate shocks, and within about 0.15 for large shocks lasting several years. At the lower bound they did not. The floor binds in one run and not in another, so adding runs missed by up to a fifth of the effect on GDP and half the funds-rate response, and shocks cannot be combined there. When the Fed’s own move is one of the shocks, read the combination as the Fed setting the funds rate that many points above where its rule would put it in response to the other shocks, for as long as the move lasts; after that the rule takes over.
FRB/US is good at what the Board’s staff built it for: tracing a shock through a whole economy whose accounts add up. Spending, production, jobs, wages and prices, the federal budget and the rest of the world all respond to one another, and a detailed financial sector — short and long rates, mortgage and corporate bond rates, stock prices and the dollar — carries monetary policy to the rest of the economy. Expectations can be switched between backward- and forward-looking, which isolates how much credibility and announcements matter. It is a mainstream benchmark, estimated on U.S. data and used by the Fed’s own staff. And it is most reliable for comparisons — one Fed rule against another, a policy paid for against one that is not, a short shock against a permanent one — and for the direction and timing of the response to a moderate shock in normal times.
The sizes are another matter. Several features of the model push its answers toward small effects, so for many shocks the numbers on this page are better read as a floor than as a central estimate.
A few features run the other way. The stock market is very sensitive to the expected path of real interest rates, so monetary shocks move equity wealth a lot (though, with a small wealth effect, much less spending). And under forward-looking expectations markets take announced policy entirely at face value, the most favorable case for forward guidance and promises to hold rates. Anything the model does not contain — industries, supply chains, a full-blown financial crisis — appears here only through its footprint on the shocks above: reallocation as a higher natural rate of unemployment, financial stress as wider credit spreads and lower stock prices.
Sources. Federal Reserve Board, FRB/US Project; the Python package and data package; Brayton, Laubach and Reifschneider, “The FRB/US Model: A Tool for Macroeconomic Policy Analysis” (FEDS Notes, April 2014); the Board’s policy rules page; the Board’s disclaimer. The model is public domain; the simulations and the shock definitions are Econvitals’.
How these simulations were built, and replication. Nothing on this page is solved while you wait: every scenario was solved in advance and stored. The library holds 13,442 solved runs—every shock at several sizes and durations, under each Fed response, with both kinds of expectations, in normal times and at the lower bound, and with both ways of paying—out of about 41,000 model solves since the Simulator began in October 2026. Rebuilding the whole library on the Federal Reserve’s November 2026 data took more than 100 hours of processor time, run overnight across two computers. The code and the Federal Reserve’s model to rebuild every stored run are in Econvitals’ public replication repository on GitHub: econvitals/replication.