Introduction section BRIDGE MODELS - Quantitative Tools for Business Cycle Analysis

BRIDGE MODELS - Quantitative Tools for Business Cycle Analysis

Bridge models are developed for short-run projections of macroeconomic aggregates and are therefore useful for tracking the Italian business cycle promptly.

These models rely on a number of monthly indicators which are readily available and allow to project macroeconomic series in a timely manner. In particular, the bridge models provide short-run projections of quarterly time series by using the information available on indicators at monthly frequency (the model thus creates a “bridge”). In general, these models are used for one- or, at most, two-quarter ahead forecasts. The approach relies on selected quantitative and qualitative monthly indicators and on the specification of a statistical equation that links these indicators with the quarterly series that has to be projected.

It is often the case that information on the indicator is not available for all months of the quarter in the forecasting horizon. In this case a preliminary work of projecting the monthly indicator is conducted so as to obtain information on the entire forecasting quarter. This is done by means of other indicators or through autoregressive models (ARIMA) projecting the monthly indicator on the basis of its past dynamics.

On the operational side, we first select the set of monthly indicators to include in the model and then ascertain whether each of them is available for all months of the forecasting horizon. If necessary, a preliminary projection of the indicator is obtained if the monthly information available does not cover the entire quarter. We then select the equation specification after ascertaining the forecasting accuracy of alternative specifications. In particular, for each equation we run rolling regressions, that is we estimate the equation on a fixed-length window of sample data that shifts forward of one quarter at a time. For each estimation on the (rolling) sample, we compute forecasts of the quarterly series (using an information set identical to that available in real time) and compare them with the corresponding actual value. This allows us to appraise the forecasting accuracy of a given model. As a benchmark, we compare the mean squared error with that obtained from forecasting with a Naïve autoregressive model.

The bridge models are employed for projecting GDP and its components on the demand side and the supply side. In the first case, we estimate each aggregate in isolation, i.e. consumption, investment, export, import and inventory change and then aggregate the corresponding projections so as to obtain a forecast for GDP.

Alternatively, GDP is estimated on the supply side and the model features two blocks. The first one refers to value added of industry and uses information on industrial production and the number of car registrations. The second one refers primarily to value added of the service sector and to a large extent uses qualitative indicators.

The monthly economic indicators we use are either of quantitative type or qualitative (i.e. when information is drawn from surveys on economic conditions). The statistical sources are the National Statistical Institute (ISTAT), Eurostat, the association of car producers (ANFIA), TERNA (for data on electricity use), ISAE and the European Commission.

Most preliminary projections of monthly indicators over the forecasting quarters are derived through ARIMA techniques. Exceptions include interest rates, calculated as implicit rates drawn from the yield curve and the exchange rates, whose expectations are obtained from the uncovered interest rate parity condition applied on US and Euro Area interest rates.

Of particular importance is the projection of the industrial production index, a key monthly indicator with large informative content in predicting the path of Italian business cycle. Given its extensive variability, we employ a variety of alternative methods for projecting industrial production and this allows us to provide different scenarios with different implications for GDP evolution.

In general, we distinguish between forecasts over a very short term (one- or two-month ahead) for which leading indicators are available (e.g., purchasing manager index, PMI and electricity use) and short-term forecasts (over an horizon up to six months). In this case we use ARIMA models for projection. In light of the importance of the industrial production index, in general we provide the battery of short-term projections of macroeconomic aggregates obtained with our bridge models in the same day when official data on industrial production are released.

Baffigi, Golinelli, Parigi (2004)Bridge models to forecast the euro area GDP International Journal of Forecasting. Volume:20

Franck Sédillot and Nigel Pain (2003) Indicator models of real GDP growth in selected OECD countries Economics Department working papers no. 364 OECD

Golinelli, Parigi, (2005) Short-Run Italian GDP Forecasting and Real-Time Data C.E.P.R. Discussion Papers 5302.

Nelson, C. R. & Siegel, A. F. (1987). Parsimonious modeling of yield curves, Journal of Business 60(4): 473—489.

Pappalardo, Rapacciuolo & Ruocco (2007) Il nuovo modello econometrico trimestrale del CSC per l’economia italiana CSC Working Paper n. 58

Rünstler G, Sédillot F (2003) Short-term estimates of euro area real GDP by means of monthly data working paper no. 276 ECB

Svensson, L. (1994). Estimating and interpreting forward interest rates: Sweden 1992-4. Discussion paper, Centre for Economic Policy Research(1051).

Related Content