OECD Composite Leading Indicator System

Ronny Nilsson and Gyorgy Gyomai

Presenting author:
Gyorgy Gyomai, born in 1975, is a statistician at Statistics Directorate OECD, Short Term Economic Statistics division. He has been working on methodological aspects of the OECD Composite Leading Indicators System since 2006. Prior to joining the OECD he worked at the central bank of Hungary as a financial markets analyst. Major areas of research that he has been involved in were yield curve estimation and term structure modelling, and market microstructure analysis with a strong focus on foreign exchange microstructure. He holds a MSc. in Economics (with a major in Finance) from the Budapest University of Economic Sciences and Public Administration, and a MSc. in Economics from University Pompeu Fabra (Barcelona) from its Graduate Program in Economics, Finance and Management (GPEFM).

The OECD System of Leading Indicators was developed in the mid-1970s in co-operation with national experts of member countries with the aim to provide early signals of turning points (peaks and troughs) between expansions and slowdowns of economic activity. This approach based on the cyclical properties of economic indicators can scarcely replace properly specified forecasting models, but it represents a relatively convenient and economic way of obtaining a broad-brush picture of aggregate economic activity from a large amount of data by construction of composite leading indicators (CLIs) for member countries.

The methodology used for the OECD Leading Indicator System has not been changed in any major aspect since the CLIs were first published in December 1981. This paper describes the indicator system and outlines the methodological changes and improvements to the system implemented during 2007. An overview of the main steps in the indicator system with indications of major changes is covered in the present paper.

The raw CLI components pass a series of filters before they are averaged to yield the composite indicator. According to the present methodology in a pre-adjustment phase they are tested for seasonality, then the series are de-trended, frequency converted if needed, smoothed and normalized. The new methodology keeps the filtering approach, however the filter sequence and some of the individual filters will be modified. The frequency conversion will be part of the pre-adjustment phase, we will test the series for missing values and outliers, the de-trending and smoothing will be performed in one step, and we will allow for lag-shifting series and weighting.

The OECD system of leading indicators is based on the "growth cycle" approach where cycles are measured in the deviation-from-trend or ratio-to-trend series. Therefore, the selection of a well behaving, de-trending method is crucial for the quality of the leading indicator. The phase-average trend method (PAT) constructed by the NBER in the U.S. and developed by the OECD is used for trend estimation. The reasons behind the intent to change the PAT method are plenty. The PAT method is not very transparent, not widely used by economists, in automated version often gives unreasonable results, in its present implementation has time series length limitations and it is using ad-hoc built-in parameters that are non-modifiable and determine the average extracted cycle length. Its non-automatic version relies on manual turning point insertion, which rather leads the method into the domain of arts than econometrics. As alternatives we consider two band pass filters: the double Hodrick-Prescott filter and the Christiano-Fitzgerald Random Walk filter. We rely on the frequency domain interpretation of the time series for the parameterization of these filters.

We reviewed also the ways we present our CLI figures for the public. As a result of this review we decided to shift the focus of our CLI related publications towards the amplitude adjusted form of the CLI (from the present trend restored CLI, and 6 month rate of change of the CLI series). We expect that this shift in presentation focus will yield clarity, and further improve comparability of our CLIs in a wide class of cyclical indicators. We also believe that this version of the indicator fits best with the intent to provide a broad-brush picture of upcoming economic activity.

Download der Langfassung als PDF-Datei