Figure 14 shows how, in the absence of further judgement, COMPASS would unwind these newly estimated shocks and thereby update the GDP forecast. The changes to the forecast depend on: first, the size of the shocks identified over the past; and second, the persistence of those shocks, which is determined by the model estimation. In this example, most of the changes to GDP growth were made to the first year of the forecast, with the level being about 0.1-0.2% higher in the medium term on account of the permanent productivity shock identified by the model, which by definition does not unwind.
Bank staff need to form a view on whether this baseline treatment is plausible. In this particular example, the treatment of business investment offers a case study of why and how one might want to use a different treatment to that suggested by the model.
Note that business investment growth over the forecast is actually stronger after the introduction of news, because COMPASS primarily attributes the new lower level of investment to a temporary shock, which then unwinds. This is most apparent in Figure 15, where quarterly growth rates average around 0.5pp more during the first half of the forecast.
In this example, one could make a strong case that the shocks which had lowered
investment over the past were likely to prove more persistent than the model treatment suggested. Equivalently, if one were sceptical about the merit of having a stronger in- vestment growth forecast in February 2012 than in November 2011, a better alternative would be to carry through the lower level of investment over the past throughout the forecast period. There are two economic arguments one could appeal to to justify that alternative treatment:
• First, the evidence from the suite of investment models at the time was not sup- portive of a stronger profile for investment growth. Figure 8 in Section 5 shows the individual model forecasts at the end of the November 2011 round, alongside the MPC’s forecast at the time. At that time, four of the seven models were suggest- ing weaker investment growth than the MPC’s forecast. In particular, the money, lending and investment system in the investment suite (described in Section 5.4), which among those models accords most weight to financial channels and spare ca- pacity, was below the MPC’s November forecast. This provides an example of how different suite models, emphasising different economic channels, can be employed as a cross-check to inform the main forecast.
• Second, the wider economic environment at the time did not suggest that a sharp recovery in investment was imminent. The euro area crisis had been intensifying in the second half of 2011 and a number of external commentators had been revising their growth forecasts for the UK downwards.130 As the February 2012 Inflation Report said, “The level of business investment is projected to recover only gradually in the near term, given the degree of slack within businesses, continuing uncertainty about future demand, and restricted access to credit”. In those circumstances staff would have been more likely to adopt a cautious approach to the processing of data news, a view that, on balance, would probably have been justified by data outturns since 2011.
Having made a decision not to take COMPASS’s treatment of the news as given, Bank staff would then have to decide how to implement the judgement to weaken investment.
One approach would be to manipulate the investment forcing process, ˆεI. This represents the cumulative total of previous shocks to investment, taking account of their decay.
Figure 17 shows how this adjustment could be done: rather than assume no impact on the level of investment by the end of the forecast, the extra shocks to investment identified over the past are assumed to persist for longer. The effect of this is to lower growth rates of investment throughout the forecast, but particularly early on.
As explained in Section 6.2.4, the MAPS inversion algorithm makes it straightforward for Bank staff to impose this particular profile for ˆεI. The desired path for ˆεI can be
‘fixed’ using a suitable shock or combination of shocks to best capture the economics of the judgement in hand. The MAPS inversion technology can then be used to determine the values of the shocks which are needed over the forecast in order to generate the desired path.
In this instance, the economics of the judgement are best captured using the invest- ment adjustment cost shock,ηI. The main reason for that is that the judgement is simply offsetting a previous change to the forecast (made when the data news was first intro- duced), and that was imposed using the investment adjustment cost shock. The standard
130See Page 49 of the November 2011Inflation Report.
76 Working Paper No. 471 May 2013
Figure 17: Applying judgement to the investment adjustment cost forcing process
2008Q2 2009Q2 2010Q2 2011Q2 2012Q2 2013Q2 2014Q2
−2.5
−2
−1.5
−1
−0.5 0 0.5 1 1.5 2
Initial news in ˆ²It
Judgemental adjustment to news in ˆ²It
Notes: The solid black line shows the change in the forcing process, ˆεI, associated with the investment adjustment cost shock generated by the revisions in past observable data. The red dashed line shows a judgemental profile for the change in the forcing process applied to lower the growth rates of business investment in the forecast.
practice would be to assume that all of these forthcoming shocks are anticipated by agents in the model.
When this judgement is applied, all other variables in the model respond to the new profile of investment adjustment cost shocks which have been incorporated into the forecast. In this context, because the balance of shocks is negative and the shock selected is a demand shock, the effect will be to reduce the level of GDP and to lower inflation over the early part of the forecast.
The GDP impact is evident from Figure 18, which shows the shock decomposition of the news in GDP following the extra adjustment. It is identical to Figure 14, except for the fact that investment shocks continue to drag on the level of GDP over the forecast.
The net effect of “introducing news” and imposing the additional judgement is that rates of GDP growth were slightly weaker at the end of this process than they had been in November, whereas under a “vanilla” COMPASS treatment with no additional intervention, they would have been slightly stronger. Note that in practice, Bank staff would repeat this treatment for each one of the shocks in COMPASS, verifying that they were content with the changes to the forecast caused by the process of introducing news.
This section has demonstrated how COMPASS and its associated tools can be used to interrogate a set of new back data and to identify the shocks that drive it. It also showed how the forecast can be updated in the light of those shocks. Most importantly, however, it emphasised the importance of judgement, including the use of the suite of models as a cross-check, in updating the forecast.
Figure 18: Shock decomposition of the news in the level of GDP, following staff judgement
2008Q2 2009Q2 2010Q2 2011Q2 2012Q2 2013Q2 2014Q2
−1
−0.5 0 0.5 1 1.5
Domestic risk premium LAP
Investment adjustment cost Export preference Import preference All other shocks
Notes: A COMPASS shock decomposition of the news in the level of GDP (measured in per cent) using the Kalman smoother, following the judgemental adjustment of the forcing process for in the investment adjustment cost shock shown in Figure 17. LAP denotes the ‘labour augmenting productivity’ shock.