Demystifying the DSGE Model

DSGE models were bashed during the Great Recession for their inability to foresee the crisis. Yet, they are still central banks’ go-to macroeconomic tool to predict business cycle fluctuations, perform policy analysis, and communicate monetary policy expectations. Despite their negative press and widespread use, these models remain less well-known outside the macro circle. Here, in Part 1, we demystify the DSGE Model. In Part 2, we cover its limitations and introduce extensions of DSGE models. Finally, we will also analyse the forecasting performance of some of these models during the turbulent times of 2008 using the New York Fed working paper, DSGE Model-Based Forecasting.

Part 1

What Is a DSGE Model?

• Dynamic – how the economy changes over time

• Stochastic – how the economy is affected by random shocks

• General – applicable to the entire economy

• Equilibrium – assumes different markets within the economy clear every period

These models are built on microeconomic foundations – how agents (consumers, producers, and governments) make an intertemporal choice (a current choice that affects what options become available in the future). These models can capture the interaction between policy actions (monetary policy) and agents’ behaviour. They can also show how random shocks within the economy may modify this interaction over time and its impact on macro variables such as output and inflation.

 

Setup

Basic DSGE models consist of three intertwined blocks: demand, supply, and monetary policy. The equations underpinning these blocks make explicit assumptions about the behaviour of the main economic agents in the economy. For example, households are assumed to optimise utility in such a way that consumption and labour effort is balanced. Firms are assumed to maximise profits by specifying the optimal amount of goods produced and the amount of labour hired.

A typical set of assumptions:

• Perfect competition exists in all markets

• All prices adjust instantaneously

• All agents are rational

• No asymmetric information (perfect information)

• Firms and households have identical preferences

 

Understanding how DSGE Replicates the Economy

• Demand and Supply Side

The demand block in Figure 1 determines real activity. This is a function of the forecasted real interest rate (the nominal rate minus expected inflation) and of expectations about productivity in the future. This block captures the connection between high real interest rates and the preference of households and firms to save rather than consume in such a scenario.

Simultaneously, agents are willing to spend more when prospects (current or future) are better. The line connecting the demand block to the supply block shows that the level of activity emerging from the demand block is a key input in the determination of inflation, together with expectations of future inflation. In good times, when the level of activity is high, firms increase wages to induce employees to work longer. Higher wages increase costs, putting pressure on prices and generating inflation. The higher that inflation is expected to be in the future, the faster the price will start to rise concurrently and contribute to a rise in inflation today.

 

Figure 1: The Basic Structure of DSGE Models

• Role of Monetary Policy

The determination of output and inflation from the demand and supply blocks sneaks into the monetary policy equation, as indicated by the dashed lines. This equation captures how the central bank sets the nominal interest rate. The central bank increases the short-term interest rate when the economy is overheating and lowers it in the presence of economic slack. By tweaking the nominal interest rate, monetary policy in turn affects real activity and this in effect impacts inflation. The solid line running from the monetary policy block to the demand block and then eventually to the supply side illustrates this.

 

• Role of expectation and central banks communication

Agents are forward-looking in these models. The way monetary policy is conducted has a significant impact on their expectations. The monetary policy of tomorrow impacts output and inflation (via the expectations channel). In DSGE models, expectations are the main transmission mechanism through which the policy affects the economy (shown by the arrows in Figure 1). The way central banks modify future expectations is via communication (take a look at our deep dive, How Central Banks Can Talk Up Interest Rates And Term Premium, to learn more).

 

• Exogenous shocks

Without any shocks, this model predicts a steady progression of the economy (no recession or booms). However, in reality, random external shocks do impinge the economic equilibrium occasionally. One example is Covid-19, which recently has negatively disrupted both demand and supply side. One strength of the DSGE model is that it allows for these shocks to be injected and allows you to track the evolution of the economy’s fluctuation as the shock transmits through different channels. For example, we can model the economic impact of Covid-19 by inputting how households’ purchase decisions (demand side) and firms’ pricing and production decisions (supply side) would be impacted (on a macro level). Based on those assumptions, we can run the model to arrive at the forecasted GDP, inflation rate, and optimal policy rate under different scenarios/assumptions over time.

 

Part 2

Limitations of the DSGE Models

The fundamental downside of DSGE models is that they ultimately become ‘jump variable systems’. For example, if a random shock hits the economy, the model would suggest the central bank would instantly hike or drop interest rates to respond. But policy is never implemented that way. On the contrary, the central bank always uses a very gradual approach, whether hiking or easing.

There are fixes to this. One was suggested by Jeffrey Fuhrer in Habit Formation in Consumption and its Implications for Monetary Policy Models. He argues that you can add habits in the form of past consumption patterns to smooth the behaviour of the model. More importantly, this lets the model incorporate more realistic behaviour of consumers who want to smooth their consumption across multiple periods.

Other limitations:

• The output of this model is sensitive to what assumptions and inputs we feed it. In modeling the Covid-19 shock in Part 1, we had to assume how agents’ consumption and production decisions are altered. In hindsight, the impact remains ambiguous and could potentially decrease the validity of the output we obtain.

• These models ignore hysteresis effects in the economy. The scale of certain shocks in the economy (i.e. GFC 2008) could be so large that it leads to structural damage. Due to this, productivity (and hence inflation) may not recover to previous trend values and could also change the dynamics underlying the demand and supply function (Figure 1). The model fails to capture this effect.

• The information available to the DSGE econometrician comprises only the data available at the time of the forecasts (usually in quarterly frequency). This means the information set used excludes the revised data (macro data is revised all the time), and the model is specified using only lagged information available on the state of the economy (it excludes the impact of time-sensitive current indicators i.e. PMI data).

• The set of assumptions used to mimic the realities of the economy are too simplistic. For example, before the 2008 financial crisis, the models ignored the impact of the financial sector on the economy. Nonetheless, extensions of these models are constantly developed to combat this.

 

Extensions of the DSGE Models

Up until now the DSGE framework we used was a closed economy, small-scale model (without a Government and Trade channel). For illustration purposes we kept it simple, however, there are a variety of extensions in the literature that could be adopted to enrich the modeling and capture the realities of the economy.

• Habit Persistence – future consumption depends on prior habits (Smets and Wouters, 2007)

• Adjustment Costs on Investments – to make investments less volatile (Smets and Wouters, 2007)

• Labour Adjustment Costs – wage stickiness (Smets and Wouters, 2007)

• Wage Bargaining and Labour Market Search (Gertler, Sala, and Trigari, 2008)

• Open-Economy Models – to account for exogenous shocks from trading partners (Adolfson et al., 2007)

• Financial Intermediation – to allow for the role of banks (GERALI et al., 2010)

• Financial Intermediation – to allow the incorporation of a central bank’s balance sheet and its unconventional policies (Sims and Wu, 2016)

• Financial Frictions – looks in liquidity constraints facing banks (Christiano, Motto and Rostagno, 2010)

• Unconventional Monetary Policy – to proxy central bank credit intermediation (Gertler and Karadi, 2011)

 

Postmortem of DSGE Models in Forecasting the Great Recession

 

Figure 2: Forecasts for Output Growth: DSGE versus Blue Chip

Source: DSGE Model-Based Forecasting

Setup

The chart compares the DSGE model with Blue Chip forecasts (a median survey result of leading economists) and the actual outcome of output growth. This is done for the three focal points during the crisis.

• 10 October 2007, when turmoil in the financial markets had just begun

• 10 July 2008, just before the default of Lehman Brothers

• 10 January 2009, at the peak of the crisis

The DSGE model’s mean forecasts are shown by the red line. The Blue Chip forecasts are in green diamonds, and the actual realisations of output are represented by the dashed black line.

The chart shows the forecasts for three different DSGE specifications. The first one (SWπ) is Smets and Wouters (2007). The second is the Smets-Wouters model with financial frictions (SWπ-FF) such as the ones introduced by Christiano, Motto, and Rostagno (2010). The final model is essentially the second model (SWπ-FF) but had the most recent quarter-end data rather than the lagged values.

 

Result

The model that is augmented with better extension to match the realities of the economy produced better forecasts. One striking result came out of the SWπ-FF-Current specification (highlighted). Incorporating the latest quarter data produced approximately the same forecast as the Blue Chip. Considering the Blue Chip forecasters can change their forecast based on new data release – this is an impressive feat for the model.

 

In a Nutshell

Think of the DSGE model as a pencil and the central banks as the poet. Without a pencil, the poet cannot write the poem (forecast economic outcome). By itself, the pencil cannot write the poem, nor it can identify whether the poem is any good (whether the forecasts are valid). It is the ingenuity of the poet (input, assumption, and extension used) that eventually determines the quality of the poem.