180 likes | 363 Views
The quest for early indicators of turning points in economic conditions: how the official statistics is involved? . Gian Paolo Oneto ISTAT International Seminar on Timeliness, Methodology and Comparability of Rapid Estimates of Economic Trends. Outline.
E N D
The quest for early indicators of turning points in economic conditions: how the official statistics is involved? Gian Paolo Oneto ISTAT International Seminar on Timeliness, Methodology and Comparability of Rapid Estimates of Economic Trends
Outline • Defining the problem at hand: what signal users are looking for? • The role of timeliness and its relationship with reliability • (a digression) An often misunderstood feature of Short-Term (ST) indicators: volatility. • The boundary between flash estimates, real time estimates (nowcasting) and forecasting: a central issue of Official Statistics (OS). • The search for leading indicators: mixing ST indicators with other sources. • A couple of possible ways forward: investing on the production of statistical indicators with leading properties; improving the quality of survey data.
1. Defining the problem at hand: what signal users are looking for? The real features of the demand of high frequency economic information that is emerging from professional users and the public at large should be clearly identified Indeed the recent developments of the economic situation are stressing the need of timely statistics able to monitor the evolution of the Business Cycle (BC). Quarterly national accounts, activity indicators, labour input, prices, foreign trade data are the target domains, and the timeliness requirements are set by the best international standards. In turn, the information provided by ST indicators is often regarded as lagging behind the “real question” that is haunting analysts and policy makers: are early signals of the recovery in sight? The crucial issue in front of large swings in economic conditions seems to be the detection “as soon as possible” of the inception of a turning point in the BC If this is the case, the problem at stake is not just timeliness: we need to understand how to channel this demand, disentangling the pressure for more timeliness (more early estimates) of ST economic indicators from the interest in indicators with leading properties.
2a. The role of timeliness and its relationship with reliability As far as the main ST indicators are concerned (Quarterly GDP, IPI, retail trade, CPI, output prices, etc.) the benefits of improving timeliness are quite clear, as shown by the Eu experience In tracking correctly the BC evolution a large set of timely indicators is needed. In this respect, the evolution of the European System of ST indicators is very interesting: assigning high priority (and reasonable resources) to this target, a good level of timeliness for key indicators has been achieved: 45 days delay for flash estimate of GDP, 40 days for IPI, 30 for industrial output prices, 45 for IP in construction, 30 days for retail trade, 0 day for CPI flash. Additional requirements are pending: the availability of timely indicators of labour market development, and indicators of service activity.
2b. The role of timeliness and its relationship with reliability Moving along the trade-off between timeliness and reliability (measured in term of revision error) is a process that must be undertaken with a clear assessment of the possible costs The information benefit of improving the timeliness of well established ST statistics (GDP, IPI) must be assessed, in terms both of feasible timeliness gains and of possible losses due to the worsening of the signal to noise ratio (larger revisions). Moreover, can timeliness help with respect to the issue of gauging early signals of turning points (or acceleration/slowdowns) of the BC? Taking as starting point the Eu situation (that is slightly less favorable for users than the one in the US) the timeliness gain does not seem the crucial factor affecting BC monitoring. The current targets (improving to 30 days the timeliness of GDP and IPI estimates, producing flash estimates for employment and LCI) can be judged as the medium-term limits to feasible improvements, unless large-scale jumps in the production process are envisaged.
2c. The role of timeliness and its relationship with reliability GDP estimates cannot be reasonably advanced further unless a substantial set of activity indicators are available at less than 30 days (in this respect the volatility of activity measures is crucial; going toward monthly estimates of GDP could be an alternative long-term strategy). As a result, I would argue that the feasible improvements (given current resources and statistical infrastructure) in timeliness can play a small role in advancing the monitoring of turning points of the BC. To be clear: to identify an upturn/downturn of the BC the last q-o-q rate of change of GDP is strictly insufficient; it must be supplemented by a quite large span (6 months?) of ST indicators (IPI, production in construction, turnover in services, employment) in order to compensate for the uncertainty (due more to volatility that to the typical revision error).
3a. An often misunderstood feature of ST indicators: volatility. Almost every ST indicator of economic activity is affected by an high degree of volatility. This feature is inherent to the underlying economic processes and/or to the measurement methods. The indicators currently utilized to monitor the BC (in particular monthly ones) have high frequency components (noise) that hide the underlying cyclical signal. This is a well known feature that in the classical literature on BC was treated with specific instruments: the MCD (Months for Cyclical Dominance) criteria was considered a kind of label to be associated to each indicator. For instance the total IPI of EMU and 3 largest Countries has an MCD ranging from 2 (for the Euro area total) to 4 (for France). The MCD increases rapidly using sector breakdowns: considering the 5 Main Industrial Groupings of the IPI index, the MCD value is never lower than 3, ranges in a large share of case between 4 and 5, going up to higher values for the energy sector.
3b. An often misunderstood feature of ST indicators: volatility. In assessing the actual timeliness of an indicator the span of monthly data that are needed to extract a meaningful signal must be taken into account. Moreover, the issue of minimising volatility must be dealt with in the phase of treating (filtering) the time series There are important implications of high volatility: the gain of high frequency indicators (monthly vs. quarterly) is low; filtering the noise component implies a loss of timeliness, as the cyclical signal emerges only averaging a long span of data. In turn, the volatility can be lowered improving the utilization of efficient procedures of Seasonal Adjustment that, indeed, are crucial to any identification of the cyclical movements of ST indicators as well as to measuring correctly their movements.
The evolution of key indicators (GDP, IPI) around a turning point of the (classical) BC in the EMU (1) Quarterly GDP and monthly IPI in levels (standardised ) at the end of the 1991/92 recession
The evolution of key indicators (GDP, IPI) around a not-so-well defined turning point of the EMU BC (2) Quarterly GDP and monthly IPI in levels (standardised ) at the end of the 2001 recession
The evolution of key indicators (GDP, IPI) around the most recent turning point of the EMU BC (3) Quarterly GDP and monthly IPI in levels (standardised ) at the end of the 2005/07 expansion
4. What are the boundaries between flash estimates, real time estimates (nowcasting) and forecasting? When OS enters the domain of flash estimates, the awkward issue of defining the limits of the estimation process must be faced. A tentative (personal) definition: flash estimate disseminated by OS are estimates based on an information set including actual data (i.e. data measured directly from respondents, via surveys or administrative files) referring at least in substantial part to the whole reference period. In other words, no part of the reference period (e.g. the last month, when quarterly data are concerned; the second part of the month, as for monthly indicators) can be measured only indirectly, through the relationship with other variables or by using projections. If the above definition holds, objects labeled “real time estimate” (usually for GDP or IPI) seem ill defined in the realm of OS, while the label nowcasting should apply to a specific typology of forecasting approaches (but should the OS be involved in forecasting?). In particular when measures of economic activity are targeted, a short delay inherently due to the measurement process has to be accepted. Collapsing this delay is tantamount to use forecasting devices based on the information set measured before the reference period: in this case how forecasts produced by OS differs from other professional forecasts?
5. The search for leading indicators: mixing ST statistics with other sources Flash estimates are a crucial instrument in monitoring the ST developments but do not seem the answer to the rising demand for indicators conveying advanced signals of turning points of the cycle. A part of the information demand points at the well established approach of Leading Economic Indicators (LEI); however this approach in recent years had not received much attention, in particular from OS. In the identification of leading indicators to be included in CLI (composite LEI), all kind of indicators can be involved: aggregate and sector specific ST indicators, money and credit variables, foreign trade data, expectations related indicators (measured by business survey data). Building and maintaining a system of LEI is valuable not only because of the forecasting properties of the CLI itself but as a structured way of analyzing the BC properties of the whole set of available indicators. Then, considering the possibility of investing in programs aimed at the identification/selection of leading indicators seems worthwhile for OS.
6a. Is there room for improving the production of ST indicators with potentially leading properties? One of the possible ways forward is to target the leading properties as one of the features that must be considered in developing (or improving) ST indicators Recently in the European Statistical System the issue of the quality of industrial new orders data and of building permits indicators has been discussed starting from the very problem that their leading properties appears to be insufficient. More analysis need to be developed about those indicators to fully understand if their limits derives from measurement practices of from the underlying behavior of economic agents. A domain where the role of leading indicators of economic activity is still very scarce is the service sector: may be there is scope for investigating the possibility of measuring demand related variables.
6b. Another possible development in the production of leading indicators: improving the quality of survey data. A key aspect that emerges from the current practices in both the leading indicators approach and the standard ST forecast modeling is the central role of survey data (business tendency and consumer sentiment data) In this respect it is time to recognize that the quality aspects of the current practices prevailing in the production of such indicators are worth to be investigated from the point of view of the OS criteria. The Ue setting, with a comprehensive program of survey data production, is already advanced in terms of comparability across countries, but could be further improved through coordination with the standard statistical production (for instance, concerning industry classification, reference years and variables to be utilized in the weighting structure). There is the perception that outside Eu the statistical underpinnings of survey data are even less comparable to the OS standards. All in all, there is the possibility that investing on the quality of survey data indicators their leading properties could be further improved.
An example of leading indicators of the IPI cycle drawn from survey data (Industry business tendency) - 1991 turning point Industrial production (IPI), Orders and Production Expectations (survey data); in levels (standardised )
An example of leading indicators of the IPI cycle drawn from survey data (Industry business tendency); 2008 turning-point Industrial production (IPI), Orders and Production Expectations (survey data); in levels (standardised )
Concluding remarks Taking as reference the situation in Eu, the feasible improvements (given current resources and statistical infrastructure) in timeliness can play a small role in advancing the monitoring of the business cycle and in particular in conveying advanced signals of turning points of the cycle. Moreover, in assessing the actual timeliness of an indicator its characteristic volatility must be taken into account (small but significant gains in the signal to noise ratio coming from adequate filtering). In turn, the information demand points at the well established approach of Leading Economic Indicators that is probably time to reconsider in the framework of OS. Then in improving the set of ST indicators priority could be given to the ones (to be identified empirically) with relevant leading properties. Finally, given the central role of survey data in anticipating BC turning points it seems worthwhile investing in their quality, possibly integrating them in the OS framework.