The Gains from Early Intervention in Europe: Fiscal Surveillance and Fiscal Planning Using Cash Data

This paper does two things. First it examines the use of real time inter-annual cash data and the role of early interventions for improving the monitoring of national fiscal policies and the correction of fiscal indiscipline. Early warnings are important because they allow us to spread the necessary adjustments over time. Examples from Germany and Italy show that large corrections are often necessary early on to make adjustments later on acceptable and to keep debt ratios from escalating. There is a credibility issue here; we find the difference between front-loaded and back-loaded adjustment schemes is likely to be vital for the time consistency of fiscal policymaking. Second, without early interventions, the later deficit reductions typically double in size – meaning governments become subject to the excessive deficit procedure and significant improvement tests more often. Thus the budget savings from early intervention and the use of cash data are significant; in our examples they are similar in size to the operating budget of the department of housing and urban development in Germany. Similar results apply in other Eurozone countries.


Introduction
The past few years, especially since the expanded debt and deficits that followed the 2007-09 financial crisis and recession, have given rise to an emerging literature on the credibility of fiscal policies and the anchoring of (fiscal) expectations. 2 It matters a great deal that expectations of future policy should be anchored; that is, fiscal decisions and their financing need to be predictable, easily understood and credible in the sense of being both plausible and likely to be sustained into the future. That being the case, it is equally important that the necessary corrections be undertaken to return fiscal policies to their intended path following a shock, slippage or data errors. Only then can we regain the full effectiveness of fiscal policy that Leeper and others 3 associate with appropriately anchored expectations.
The goal of this paper is to show how to use cash data (up to the minute, in real time, and always available) on the state of government finances to help us assess fiscal developments and issue "early warning" signals about the corrections that need to be made in order to bring public finances (both deficits and debt) back on their intended track -as may be required by the stability and growth pact (SGP) from time to time. We show that early warning signals can be highly valuable because they allow us to start an adjustment process earlier and spread the effort over a longer time period. That makes the size of each adjustment smaller and a less contentious issue; and therefore less likely to create opposition in a government or its electorate. We also examine whether the budget corrections are better frontloaded or back-loaded -we find that different adjustment mechanisms have different size and timing profiles. These are all relevant questions from a policy perspective. It is important that the plans submitted or announced are actually time-consistent, so that governments stick to their published targets after an unplanned departure from the officially agreed stability programme (SP). 4 We highlight this point because there seems to be considerable evidence that the time consistency in SP plans has been violated in a number of cases.
It has long been emphasized in the fiscal forecasting literature, that data on government deficits are only available with a long time lag. 5 At the time when the data becomes available, the deficit has already realized and no policy measures can be implemented to avoid drifting away from the intended target values. However, Pérez (2007) and  and Onorante et al. (2008) show that intra-annual data, available with much shorter time lags, can be used to derive accurate forecasts for end-of-year fiscal outcomes. We first build on their results by introducing cash data into that process. We then go one step further and ask, using their forecasting model, how much a government can actually gain by using early warning signals from this procedure and engaging in early corrections of a fiscal slippage. This question has received little attention in discussions on the prudential surveillance of fiscal policy. The reasons for that are twofold. On the one hand, the question requires an objective function for the government in order to relate departures from a proposed plan to the adjustment costs inherent in making a correction. We discuss that issue in this paper and provide some alternative measures. On the other hand, it raises the problem of time consistency in fiscal policy. We highlight the potential for time inconsistency, and discuss the trade-off between adjustment costs and time inconsistency. At that stage we need to draw out the connection between deficit and debt surveillance within the SGP rules. Both parts need to be monitored if we are to fulfil the purpose behind the SGP -and create fiscal policies that are sound in the long run.
We have set our analysis within the framework of the existing SGP because that is still the standard framework for monitoring and control of fiscal policies in the Euro-zone. However the importance of our general approach has been reinforced by the fiscal crises of 2009-10 where programmes of exact deficit and debt targets, and the use of cash rather than accruals data, have been imposed on the delinquent countries -even if the parameters and time scales have had to be adapted to fit the circumstances. Similarly, our distinction between additive and slope correction schemes (ie between frontloading and back-loading the deficit corrections) has been pushed to centre stage now that preventing any further debt escalations has necessarily become the prime focus of attention. Backloading, as we show, allows debt ratios to rise slowly but surely and have had to be ruled out for that reason. Whether this will lead to a formalisation of debt target rules within the SGP remains to be seen. But these developments all serve to highlight the importance of including both early interventions and cash data monitoring in any realistic fiscal surveillance programme.

Data
The data we use in this paper come from three different data series. First, since we are interested in the deficit of the general government, we use the annual series of government deficits according to ESA95 accounting standards. We then augment this series with data collected using the ESA79 standards in order to construct a longer time series. This data comes at annual frequency and provides the official deficit figure when it comes to the assessment of compliance with the SP plans. Second, starting in 1995, there is also a series at quarterly frequency of government deficit data. This also follows ESA95 accounting standards. We use this data as intra-annual information to forecast the end-of-year government deficit. Finally, we use the data series of the public accounts (cash series) that comes at monthly intervals. In addition, we use quarterly nominal GDP data. These data are required to normalize all other variables as ratios of GDP. We decided to build the model in ratio form in order to avoid the problem of having to forecast deficits and GDP separately to generate deficit ratios for the current year which are comparable to those submitted in the SP plan.

ECB Working Paper Series No 1220
July 2010 Thus the data we use come at three different frequencies. We can reduce the frequency of the cash data by aggregating it to quarterly observations. How we deal with the remaining frequency mismatch is explained below, in the model and estimation sections.
In Figure 1 below, we plot the annualized ratio variables for the public (cash data) and national accounts (ESA95) deficits as ratios to GDP. These are the data series to be used in the estimation and forecasting procedures described next. Figure 1 shows that these data display a joint long run trend, and strongly correlated short run co-movements. This suggests that the information obtained from the cash data deficits is likely to improve the forecast performance for the annual ESA95 deficit. The increase in forecast performance from using cash data and quarterly ESA95 data has been studied extensively in Onorante et al. (2008) and Pedregal and Pérez (2008). As described in Onorante et al (2008), this finding is notwithstanding some differences in the accounting rules and conventions between the cash data and ESA95 data. These differences relate to the methods for compiling the data, timing of recording of transactions, and differences in the coverage of budgets between countries and over time.

Estimation and forecasting
To forecast the fiscal developments in the different countries, we build on the approach using intra-annual cash data developed in Onorante et al. (2008) and Pedregal and Pérez (2008). Since the SP plans submitted by the governments are always expressed as ratios to GDP, we express the cash and deficit data as ratios to GDP. To obtain intra-annual GDP ratios, we construct quarterly GDP ratios as follows: where x is the variable to be normalized 6 , y is nominal GDP, q is the quarter index and t the index for the current year. In the fourth quarter the sums comprise all four quarters of the current year and the ratio variable coincides with the annual ratio of that variable. The variables used in the estimation process are the cash-to-GDP ratios and quarterly deficit ratios. These ratios will potentially show some cyclical pattern and we control for that in our estimation procedure (section 3.2). Lastly, when we make forecasts, we assume that data is available up to and including the first quarter of the current year for the deficit, cash data, and GDP data.
As noted, the reason we cast the model at quarterly frequency is to avoid the problem of forecasting both GDP and the deficit to construct deficit ratios for comparison with the numbers submitted in the SP plan. However, this way some cash information will be not used in the current forecasts when the current GDP data is not yet available.

Model specification
Once we have constructed all the time series in ratio form, we use the state space model described in detail in Onorante et al. (2008). This model combines the mixed-frequency data from annual ESA95 publications and from the monthly and quarterly public accounts. The annual frequency is reflected in an error-correction model, specified in ratios to annual nominal GDP. The co-integrating relationship underlying the errorcorrection model exists between the indicators derived from the annual fiscal data in cash accounts (sum of twelve months within the same year), and the actual annual fiscal data. Onorante et al. show that such co-integrating relationships between the deficit, revenue and expenditure ratios 7 on the one hand and the cash figure indicators on the other hand exist in almost all cases considered. In addition, it is shown that the cash-based data is a valid leading indicator for the annual fiscal variable. The error-correction model is then converted into state-space form with quarterly frequency, using the annual ratio variable in the fourth and filling in missing values in the first three quarters of the year.
Thus the monthly intra-annual data is transformed into quarterly data, such that the intraannual model can be specified in quarterly frequency. The model for the indicator variable is directly set up in state-space form. It is thus possible to combine the annual error-correction model and the quarterly model for the indicator variable and set up a ECB Working Paper Series No 1220 July 2010 joint framework such that the output of the indicator model is incorporated as input to the ECM equations.

Detailed Forecast Results
In order to be able to exploit the model's forecasts, the forecasts should be made as early as possible but also be as accurate and reliable as possible. Facing these two conflicting requirements, we consider the mid-point of each year to be an appropriate choice for the forecasting origin. 8 At that point, we can use all information up to and including Q1 of the current year as argued in Onorante et al. (2008). The model yields the forecasts given in table 1. The row forecast reports the forecasted value by the model, the row realization gives the true realization, and the row SP gives the most recent SP plan reported by the government.  Table 1 shows the model works rather well in detecting financial slippage and therefore confirms the results obtained in Onorante et al. (2008). Although there are a few significant deviations in absolute terms between forecasted values and the realized deficits, the qualitative predictions are very good and justify the use of the model as an early warning tool -at least for the Euro-zone countries. Increasing the quantitative performance of the model is beyond the scope and intention of this paper and is therefore left as a subject for future research. 9

Timeline of events
Our goal is to study the gains that can be obtained from early intervention in fiscal policy by governments. To do that, we have to specify the sequence of events of the government's action. In period t−1, before the current budget year starts, the government submits a SP plan to the European Commission. Then at the start of quarter Q3 of year t, the first year covered by the SP plan, a deficit forecast for the entire year is made 10 . If the forecast predicts a slippage of the government deficit below the SP target, the government can consider putting early intervention measures into place immediately, in order to avoid a deficit realization that is too far away from their SP target. Independent of whether early intervention measures have been implemented or not, the government also has to adjust its plan for the three remaining years of the plan to satisfy the debt targets of the initial SP plan by the end of the plan period. We illustrate the timing of these events in figure 2: The time line set out in figure 2 is conceptually straightforward, but perhaps difficult to implement from a policy perspective. It puts a lot of faith in the effectiveness of fiscal activism, and in the "costless" production and pursuit of SP plans. Many governments may find that hard going in practice, and that mid-year is a bit late to adjust current (and future) budget plans which may have been in preparation for some time. Nevertheless, ECB Working Paper Series No 1220 July 2010 using real time data, Beetsma and Guiliodori (2008) show there is considerable evidence that governments do in fact adjust their budgets mid-year. We may therefore assume that the time line in figure 2 is feasible, if uncomfortable, at least for governments that are sufficiently determined to correct any budgetary failures before they run out of control.
That is not to say that all governments do make fiscal corrections in good time. Recent experience in the Euro-zone shows very clearly that, while some do, many do not.
In an extension to figure 2's timeline, we also discuss the case where no changes to the original plan are made and calculate the deviations in the debt ratio that then appear at the end of the SP plan period.

Correction plans
When there has been a one time deviation in fiscal policy such that the deficit exceeds the SP plan, there are at least two ways to correct the deficit and meet the debt target at the end of the SP. One method we label the constant slope plan, and the other an additive constant plan 11 . The additive constant plan takes the original plan and adds a constant to all deficits planned under the original SP agreement. This constant d therefore describes a level shift. This plan will have large adjustments between Q3 of period t, when the corrections start if there are early interventions (or from the end of period t if not), and period t+1 when the corrections can continue at the rate originally planned. Either way, these adjustments will be enough to eliminate the extra debt caused by the deficit slippage. We sketch this correction process in Figure 3.
The constant slope plan does not have this front-loading property. Instead, it creates a constant change in the deficit each year. The burden of adjustment in this plan is therefore spread more equally over the remaining periods of the plan. This time, the constant d describes a constant change in the deficit from year to year, whereas d in the additive case described a constant shift in the deficit compared to the original SP plan. We sketch this correction plan in figure 4.

ECB
Working Paper Series No 1220 July 2010 Note that the SP path is identical in both cases, so that we can easily compare the correction paths (see figure 5). We see that the constant slope plan has much smaller corrections in the deficit in the periods t to t+1. The larger corrections are postponed to later periods where, compared to the additive case, the adjustments relative to the original SP plan are plainly larger. They are back-loaded therefore 12 . In that sense these two correction schemes are polar opposites in terms of where the adjustments fall. The frontloaded, additive constant case represents a rapid correction ("cold shower") strategy and risks stirring up political opposition. The back-loaded, constant slope version is easier to implement, but risks allowing the policymakers to backslide and will therefore lack credibility if the markets doubt that it will be carried through to the end.  14 ECB Working Paper Series No 1220 July 2010

Example 1: The case of Germany in 2002
To give concrete examples of the differences between the two correction schemes we have described, we first consider the case of German fiscal policy in 2002. The German government submitted the following plan (see  For this plan, we take the deficit numbers and the nominal growth rates as given. The debt level in 2001 is taken to be 60 % of GDP. We update the debt ratio according to the following rule: where D denotes the debt ratio at the end of period t, t is the deficit in period t, and N t g is the nominal growth rate of GDP between periods t and t−1. 13 The deficit t Δ contains interest payments on the accumulated stock of debt For the purposes of illustration, we abstract from changes in the interest rates over time in this paper. It is however straightforward to extract the interest payments component and analyse it separately from the primary deficit. Based on the forecasts made for Germany in June/July 2002, we expect a deficit for 2002 of 3.88%. 15 We now derive the actual debt ratio of the German government using the realization for 2002 and assume that this is a one time deviation from the SP plan (see table 3). This case is constructed to resemble a situation in which no intra-annual interventions are made, and the full deficit is realized at the end of 2002.If the 13 t Δ is negative if the government runs a budget deficit, and positive if it runs a surplus. The planned debt ratios correspond to the debt ratios submitted by the German government, except for 2005 where the government reported a debt ratio of 55.5% but the updating procedure gives a ratio of 54.55%. 14 i denotes the nominal interest rate on debt from period t−1 to t. 15 It will turn out that, due to favourable shocks, the actual deficit was 3.66% for 2002.

ECB
This step is used to provide a start to our calculations for the additive and constant slope corrections with early interventions. We replace it with a less arbitrary, but rather more difficult to implement, assumption for the early interventions in section 6.3. By contrast, where there are no early interventions we start with the realised deficit at the end of period t.

The additive correction approach
For the correction of a one time deficit slippage, we look first at the approach with a level shift for the remaining years up to the end of the planning horizon, in this case to 2005. The level shift is chosen such that the government can still satisfy its originally announced debt target. 16  It is now easy to solve for d using the law of motion for the debt ratio in (2)   interventions, plus the additional interest charges. By contrast, the 0.83% difference in the early interventions case has to be corrected through public spending cuts in 2002. 17 That may be a tough call, but it does allow the German government to avoid the SGP's excessive deficit procedure and it is the kind of correction the German government will impose upon itself in its own new balanced budget legislation.

The slope correction approach
The second approach for correcting the one time deficit slippage would be to apply a constant slope adjustment path for the remaining years up to the end of the planning horizon, in this case 2005. The slope has to be chosen such that the government can still satisfy its original debt target. We need to find deficits Given that, it is easy to solve for d based on the law of motion for the debt ratio in (2

Altering the impact of early intervention
For our numerical example, we assume that, at the point at which the slippage is first detected, it is still possible to achieve a deficit at the end of the year that is a convex combination of the planned and the realized deficit. To overcome possible concerns regarding this particular approach, we provide a flexible sensitivity formula. Instead of assuming equal weights, we tried introducing a weighting parameter θ such that where t Δ denotes the realized deficit in period t without intervention. It is easy to verify that the adjustment rules for d under both rules are linear in θ . The derivative with respect to θ therefore provides a flexible sensitivity measure of the adjustment rule with respect to the assumption of equal weights.

ECB Working Paper Series No 1220
July 2010

Constant slope adjustment
In the case of constant slope adjustments, the derivative is complete correction as soon as the slippage is first detected, we would get a slope coefficient of 1.85 % which is almost exactly equal to the numerical value of 1.83 % used earlier. For the numeric example in this section, we assume that for the remaining six months of the year after the forecast has been made, the government can implement policy measures, e.g. a budget freeze, so that the deficit is in accordance with the submitted SP plan. In the case of Germany, every month less of compliance with the SP plan 18 will increase the slope coefficient by approximately 0.09. This shows that there are quantitative effects from altering the impact range of early intervention but, compared to not intervening, the gains always remain substantial.

Additive constant adjustment
In the case of the additive constant adjustment the derivative becomes

Example 2: The case of Italy in 2002
As a second case, we take Italy in 2002. The Italian government had submitted the following plan (see  For this plan, we again take the deficit numbers and the nominal growth rates as given. The debt level in 2001 is taken to be 107.5 % of GDP, and we update the debt ratio according to the rule given in (2). 19 Based on the forecast for Italy made in June/July 2002, we expect a deficit for 2002 of 3.88%. 20 We now derive the debt ratio of the Italian government using the realization for 2002 and assume that this is a one time deviation from the SP plan (see table 8). This is intended to resemble a situation in which there are no intra-annual interventions and the full deficit is realized.
Next, we consider the situation where, after the forecast is made, early intervention measures are taken. We make the same assumption as before; that the measures taken are effective to the point that, at the end-of-year, the deficit is the convex combination with equal weights of the SP target and the realization for that year. In that case, the deficit for    The alternative would be not to take any early intervention measures and to let the deficit take its course for 2002. In that case the debt at the end of 2002 would be 105.63%. Based on this, we can calculate the level shift, d, that would be necessary to reach the 89.78% debt target in 2005. We get d= 0.0076. This implies that the government has to double its deficit reductions, to 0.76% of GDP, each year for three years, instead of only 0.38% had early intervention measures been taken. However, it must be kept in mind that the government already had to reduce the deficit by a large margin in 2002 in the early interventions scenario. The deficit in 2002 with early interventions is 1.76%, whereas it is 3.01% without such interventions; the difference of 1.25% being spread over the remaining three years. On the other hand, the deficit reductions of 0.38% for three years in the early interventions case are less than the 1.

The slope correction approach
The second approach is the constant slope adjustment path taken to the end of the plan's horizon. The new plan can be seen in table 10. The back-loading is more obvious here than in the German example; and, with an extra 1% of GDP in budget cuts in 2005, and a correspondingly larger gain in frontloading relief, the temptation for time inconsistent behaviour compared to the early intervention solution in table 8 will be large. This is guaranteed to raise doubts about the credibility of this as a budget consolidation plan.

Measuring the gains from early intervention
It is relatively easy to derive a monetary measure of the gains to be had from early interventions. Here we use the difference in the debt burden at the end of the current SP plan starting with or without early interventions. That takes out any differences in the subsequent fiscal corrections which depend on the type of correction method chosen.

ECB Working Paper Series No 1220
July 2010

The gains in monetary terms
a) For Germany: we assume that, after the slippage, the government returns to the fiscal contractions in its original SP plan, first having made an early intervention and then without. The difference in outcomes is easily read off from

The gains in terms of budget management
In the case of the additive correction approach, the gains from early intervention can be read off from the adjustment factor d that captures the value of reducing the fiscal contractions in the following years of the SP plan. In the constant slope approach, the change in the fiscal impulse can be derived as the difference of the corrected plan and the original SP plan.
We set out the net fiscal contractions or expansions implied for Germany, relative to the original SP plan, in table 12. These figures show that the two approaches offer different schedules for correcting fiscal slippages. Both allow quite a large amount of extra spending in the first year of the plan. But then they differ. The constant slope plan offers two additional years of extra spending, but then demands a large cut in net spending. The additive constant approach, by contrast, offers small cuts spread evenly over the three correction years.
For Italy, the net fiscal contractions or expansions implied, relative to her original SP plan are set out in Table 13. Qualitatively the results resemble those for Germany. They imply small but equally spread fiscal contractions under the additive approach; and but positive impulse in the first two periods, followed by a small contraction in 2004 and a sharp contraction in the last period under the constant slope approach.

Elsewhere in the Eurozone
Up to now, we have looked only at two countries in two specific time periods. In this section we do the same analysis for the eurozone countries in all years for which SP plans exist and are publicly available. To get a comprehensive measure, and to avoid an overload of individual country detail, we average the fiscal impulses of the different adjustment plans over all the cases back to 2000 where forecasting deficit slippages demanded early intervention. The results are presented in table 14 for both additive and constant slope approaches.
These results confirm that what we saw above for Germany and Italy in their 2002 plans does in fact hold more generally. The differences between the left and right panels show that there are clear and substantial advantages in terms of both lower debt levels and permanent interest savings in every case to instituting early corrections. The changes in the overall run of fiscal policies required to make the necessary budget corrections is almost always smaller -often substantially smaller -with early interventions; most obviously Germany, Italy, the Netherlands and Belgium. In most cases the changes are halved if early action is taken; with those changes bunched in the final year in the slope adjustment method is used, but evenly spread out if the additive method is used.  Similarly, the required changes are almost always much smaller (even in the earlier years) if the additive adjustment method is used. In fact the changes are typically 3 to 5 times smaller with additive rather than slope adjustments, most notably in Spain, Germany, Italy and Portugal (possibly Belgium and the Netherlands). It is important to bear in mind that, although these results are qualitatively the same as those for the actual budget corrections discussed in sections 6 and 7, the figures in   Looking at the figures in tables 15 and 16, we can see that the interest payment consequences of making no early budget corrections are indeed small (justifying our having left them out of consideration before this point) -just a fraction of a percent in each case, even in Italy and the Netherlands. Moreover the extra interest payments are not different between the two correction methods. Again it was reasonable to ignore them. However, they build up over time and last for a long time since the corrections are sufficient only to get us back to the planned debt levels.

Conclusions
In this paper we have accomplished two things: a) We have investigated the gains to be made, in terms of monitoring and correcting excessive deficits, by using cash data rather than accruals data to monitor the current fiscal position; and, given that information, by computing the early interventions needed to head off any excess deficits as they emerge; and b) We have also examined and compared two different strategies for correcting excess deficits that may have already emerged, such that the debt ratio returns to some prespecified level at the end of the designated planning period.
Independently of the measure used, our results obtained show that the gains from early intervention are certainly not negligible for governments committed to reducing their debt. What is different between the different plans is the allocation of the corrections over time. Both the constant slope and the additive constant plans require a lot of effort at the moment a fiscal slippage is detected in order to avoid further slippages in the remainder of the year and in subsequent years. This means that an important part of the adjustment has to take place in the first year. After that, the two approaches differ sharply in the timing of the corrections. The additive approach typically requires a second costly adjustment after the early intervention. One could however argue that this would be a less serious problem if those corrections could be implemented in a package together with the early intervention measures and therefore avoid further discussion in the political process.
The advantage of the constant slope approach is that it spreads the adjustment steps equally across the horizon of the SP plan. They may therefore be easier to implement ECB Working Paper Series No 1220 July 2010 politically. However, it also requires an especially strong and committed government to overcome the (political) temptation to use the surplus that occurs towards the end of the plan for opportunistic spending instead of debt reductions; for example, close to elections or as debt or deficit consolidation fatigue begins to set in among the politicians.
In our discussion of the different adjustment paths available to a fiscally consolidating government, we have emphasised the gains of early intervention. In that context we have stressed the importance of accounting for the interaction between deficit and debt targets, and the danger of time consistency in the associated consolidations. The time consistency problem in particular seems to be empirically relevant and strategically important, but it has received very little attention in the literature on the monitoring and control of fiscal policies.