Can we systematically improve upon using preliminary vintages of economic data?

The problem
Much economic data is not what it seems, especially data that informs economic policy! A leading example comes from the National Accounts data produced in the UK by the Office for National Statistics (but all industrialised countries have the same problem). Consider data on for example GDP, which is critical to an assessment of policy, this is published with a lag of about one quarter, so that policy can be based on data with a ‘reasonable’ delay. However, complex data collection systems base such initial estimates on incomplete information: as more information is collected data that has already been published is subject to changes; for example what was a ‘double-dip’ recession in GDP, may well no longer be on the newly informed data!
This research proposal offers a solution to this conundrum. Put simply: forecast the ‘true’ data using the earlier data. However, whilst simple in concept, there are naturally complications in practice. Most approaches to date are ‘model bound’. That is they choose a particular paradigm of model building and then apply that, quite often with little or no assessment of whether the framework is the best available. In this project, there will be an explicit relaxation of the modelling framework to invoke a minimal set of assumptions. There is good, although presently somewhat limited, evidence that this approach can outperform typical model-based frameworks. This project will explore this issue and develop the flexible framework in a systematic and comprehensive manner, which will offer gains to policy-makers, whilst also meeting the demands for peer-reviewed publications.
Theoretical Development
The key question is whether it is possible to improve upon the preliminary vintages in order to provide more accurate forecasts of subsequent vintages? In this study we construct forecasts of the final vintage by different methods and evaluate them against baseline of the preliminary vintages. This problem also raises a relevant question: whether the first available (or more generally preliminary) vintage(s) of the data can be considered as noise or news. Noise suggests that as the different stages in the revision process progress, there will be convergence to the ‘true’ values of the variables being measured; ‘news’ suggests that revisions are in a sense a ‘surprise’, they are not then predictable given information prior to their occurrence.
Applications
An ambiguous, but achievable, application in this context would be to link multiple revisions and multiple variables, thus exploiting possible data revision relations between variables, for example, consumption and income or GDP and industrial production. Such an extension necessarily leads to an increase in the dimension of the system being considered, but offers gains by exploiting the correlations between revisions and between variables. A further extension would be to consider the policy-based implications of better forecasts of vintage-based data, for example in assessing the public expenditure implications of different (vintage) estimates of GDP growth