Stress-testing the Impact of an Italian Growth Shock using Structural Scenarios

Transcription

1 Stress-testing the Impact of an Italian Growth Shock using Structural Scenarios Juan Antolín-Díaz Fulcrum Asset Management Ivan Petrella Warwick Business School June 4, 218 Juan F. Rubio-Ramírez Emory University Introduction In recent weeks financial markets have suddenly awoken to the economic and political risks posed by a populist coalition government in Italy. From May 9 th, 218 to May 29 th, the yield of two-year Italian government debt jumped from.13% to 2.74%, an increase of almost 2.9%, the most rapid jump ever recorded in the space of 2 days. These developments have raised the question of what could be the fall-out of Italian political developments to the rest of the euro area and the world, were Italy to fall into an economic recession. In this short note, we apply the structural scenario analysis tools that we have developed in a recent technical paper 1 to stress-test the impact of an Italian growth shock on economic activity in the main other European economies and the United States. Modeling Italian Growth and its Spillovers There are two main challenges to construct a valid scenario for stress-testing the impact of Italian growth on other countries. The first challenge relates to measuring growth itself. The most basic approach would be to use real GDP growth rates for the countries of interest. However, GDP growth Corresponding author: Juan Antolin-Diaz <juan.antolin-diaz@fulcrumasset.com>, Department of Macroeconomic Research, Fulcrum Asset Management LLP, 66 Seymour Street, London W1H 5BT. 1 See Antolin-Diaz et al. (218) and the references cited therein for a description of the methodology.

2 % Annual Growth Stress-Testing Italian Growth Figure 1: Italy Unconditional Predictive Density and 5 th Percentile Note: The solid black line is the indicator of economic activity for Italy. The fan chart represents the predictive distribution based on the forecast of the BVAR described in the main text. The areas plotted are, from lighter to darker, the 99%, 9%, 8% and 68% confidence bands. The dotted red line highlights the lowest 5 th percentile of the predictive distribution. is noisy and contaminated by short-term measurement error, which could dampen the correlation between countries. Moreover, GDP is only available quarterly, whereas, as we will see below, a monthly indicator is more appropriate to identify an Italian-specific growth shock. For these reasons, we use as a measure of growth for each country the monthly economic activity factor extracted from a panel of real activity variables using a Dynamic Factor Model. 2 We estimate factors for Italy, Germany, France, Spain and the United States and, in a second step, embed them into a Bayesian Vector Autoregression Model. 3 The latter model is ideal to study the dynamic interrelations between the five economies, as it captures dynamic correlations between the variables. Figure 1 plots the Italian economic activity indicator (black line) together with the predictive distribution (blue fan chart) for the next two years coming from the BVAR model. As we can see, after three years of growth, Italian economic activity has slowed down to close to zero, which is 2 The specific model used for the analysis is the one developed by Antolin-Diaz et al. (217) and maintained at Fulcrum to produce our regular nowcasts. 3 We use a BVAR(12) with a Minnesota-style prior (see Doan et al. (1983)). By using the estimated factors as endogenous variables, we are neglecting the uncertainty related to the fact that the factors are estimated regressors. Ideally, one would want do conduct the exercise in a single step, but this would require estimating a very large factor-augmented VAR, which we leave for further research. 2

3 in fact the long-run growth rate of Italy estimated by our model. The fan chart projects that the economy will continue around its zero trend growth for the next two years. The dotted red line highlights the lowest 5 th percentile of the predictive distribution. In other words, this is a quantification of an adverse outcome for the Italian economy, as emphasized in the recent work by the IMF on growth-at-risk in analogy to the Value-at-Risk (VaR) concept in financial stress-testing. 4 Identifying an Italian-Specific Growth Shock The second important challenge facing our exercise is one of distinguishing cause and effect. The Italian economy is highly interconnected with the euro area, and in turn with other areas of the world. Traditional approaches to conditional forecasting exercises rely exclusively on correlations. 5 As we highlighted in our aforementioned paper, for the present case this risks exaggerating the impact of an Italian recession on other countries economic activity: it might be that the simulations are capturing the likelihood that if Italy is in recession, it is being caused by a crisis abroad spilling over to Italy and not the other way round. To solve this problem, it is important to specify that the Italian recession needs to be caused by an Italian-specific growth shock, and rule out external shocks. We use a simple method to define such a shock: an Italian-specific growth shock is a shock that affects Italian activity immediately but has zero impact on the other countries on the first month, and can be transmitted to the other countries after the second month. 6 While this type of zero-impact restrictions have been criticized for their stringency (it is conceivable that an Italian shock could transmit to Germany and France in the space of days or weeks), it is a simple and tractable approach that helps us define Italian-specific shocks without the need to introduce additional variables. The use of the monthly indicators of activity, rather than quarterly GDP growth, makes the identifying assumption more credible. Note as well by defining the shock in that way, we are ruling out a generalized euro area confidence shock. Rather, the shock we have in mind is originating in Italy and only leading to spillovers 4 See Adrian (218), and Adrian et al. (forthcoming) for more details about the growth-at-risk approach. 5 The classic conditional forecasting methodology for VARs was developed by Waggoner and Zha (1999). See also Banbura et al. (215). In Antolin-Diaz et al. (218) we extend these methods to a general class of structurally identified VARs, including sign, zero and narrative sign restrictions. 6 This is achieved by using the well-known Choleski decomposition with Italy ordered last in the vector of observables. 3

4 Figure 2: Structural Impulse Responses to 1 SD Italian Shock.2 ITALY to Italian Shock.2 GERMANY to Italian Shock Months Months.2 FRANCE to Italian Shock.15 SPAIN to Italian Shock Months Months.3 USA to Italian Shock Months Note: Each panel represents the response of the country s indicator of real economic activity, expressed in units of annualized GDP growth, to a one standard deviation Italian-specific growth shock. The bands represent 68% and 9% pointwise credible sets. 4

5 from the second month on. Figure 2 displays the estimated impact of such a shock through the Impulse Response Functions (IRFs). Since the activity factors are in the same units as annualized GDP growth, the numbers can be interpreted as response of annualized GDP growth to a one standard deviation shock. As can be seen from the first panel, the typical Italian-specific shock affects economic activity by.25 percentage points of GDP growth, builds over the first few months to up to almost.4 percentage points and dissipates after about one and a half years. The other panels display the responses of other countries. The impact is, by assumption, zero the first month, and the response builds up and typically peaks around 6 months for all economies. the response is noticeably larger for Germany, which at peak impact responds by.2 percentage points, is intermediate for France, and smaller for Spain and the US. Interestingly, in the case of the US the initial response to the shock is subsequently followed by a strong response of the opposite sign in the second year, implying that the effect on the level of GDP is eventually re-couped. Applying the Stress-Test Based on the results above, we now construct the structural scenario for the stress test. Figure 3 displays the results. The scenario is defined by setting Italian real activity growth to its 5 th percentile outcome for one year, i.e. a recession that lowers GDP growth to about -2% in annual rate. The assumed path is displayed in red in the first panel of Figure 3. The model then calculates the sequence of Italian-specific shocks required to generate such a path, and applies it to all countries. It turns out a 1.7 s.d. shock is required on impact, followed by a succession of additional negative shocks for the entire year. 7 The result of the exercise for the other countries, as well as for Italy in the second year, is displayed as the blue line (median) and 68% bands. For comparison, the unconditional forecast (i.e. the expectation of the model in the absence of further shocks) is displayed by the solid gray lines. As can be seen, the median response is a slowdown in all countries. It is most meaningful for Germany, which would go into recession with more than 5% probability, less pronounced in France and Spain, which would still experience a meaningful decline in growth, and relatively mild for the United States, where the scenario is not significantly different from the 7 One might argue that the required sequence is unrealistic. In our paper we provide tools to assess the plausibility of a given scenario. 5

6 baseline path. We therefore conclude that Germany s economy is the most exposed to an Italian growth shock. We find this result intuitive, given the great amount of linkages between the two economies. 6

7 % Annual Growth % Annual Growth % Annual Growth Stress-Testing Italian Growth Figure 3: Structural Scenario: 5 th Percentile Italian-Specific Growth Shock 4 Italy 1 Germany France 6 Spain United States Note: For each panel, the solid black line is the indicator of economic activity for the country. The dotted black line is the median forecast in the baseline scenario. The red line for Italy is the assumed stress-test. The fan chart represents the predictive distribution based on the structural scenario. The areas plotted are, from lighter to darker, the 9%, 8% and 68% confidence bands. 7

8 Quantifying the impact on stock markets Finally, we take a step to assess the impact of the scenario on stock markets. To do so, we take the estimated series of shocks consistent with the scenario and project it on the monthly returns of different stock markets using a linear regression. This exercise ignores the uncertainty stemming from the fact that the shock series itself is an estimate, and therefore should be taken as indicative only. The results are displayed in Table 1. The first column reports the estimated regression coefficient, and the second column reports the cumulative impact of the scenario, computed as the sum of the shocks over the horizon of the scenario multiplied by the coefficient. As we can see, the shock would generate about a 1% cumulative decline in the Italian stock market (FTSE MIB), about % declines in both the European index and the German DAX index, and a smaller but still substantial decline of 5% on the S&P 5 index. Table 1: Estimated Impact of Italian Recession Impact of 1 s.d. shock Cumulative Impact of Scenario on Stock Markets FTSE MIB % DAX % EUROSTOXX % S&P % Conclusion In this note, we have used recently developed methods in econometrics such as Dynamic Factor Models and Bayesian Vector Autoregressions to construct a stress-test to an Italian-Specific shock using structural scenario analysis as advocated by Antolin-Diaz et al. (218). We conclude that the euro area, and Germany in particular, would be importantly exposed to a shock to Italian growth. 8

9 References Adrian, T. (218): The Growth-at-Risk Approach to Assessing Global Financial Stability, Technical note, International Monetary Fund. Adrian, T., N. Boyarchenko, and D. Giannone (forthcoming): Vulnerable growth, American Economic Review. Antolin-Diaz, J., T. Drechsel, and I. Petrella (217): Tracking the slowdown in long-run GDP growth, Review of Economics and Statistics, 99, Antolin-Diaz, J., I. Petrella, and J. Rubio Ramírez (218): Structural Scenario Analysis with SVARs, Discussion Paper 1272, Centre for Economic Policy Research. Banbura, M., D. Giannone, and M. Lenza (215): Conditional forecasts and scenario analysis with vector autoregressions for large cross-sections, International Journal of Forecasting, 31, Doan, T., R. B. Litterman, and C. A. Sims (1983): Forecasting and Conditional Projection Using Realistic Prior Distributions, NBER Working Papers 122, National Bureau of Economic Research, Inc. Waggoner, D. F. and T. Zha (1999): Conditional Forecasts in Dynamic Multivariate Models, Review of Economics and Statistics, 81,

10 Disclaimer This material is for your information only and is not intended to be used by anyone other than you. It is directed at professional clients and eligible counterparties only and is not intended for retail clients. The information contained herein should not be regarded as an offer to sell or as a solicitation of an offer to buy any financial products, including an interest in a fund, or an official confirmation of any transaction. Any such offer or solicitation will be made to qualified investors only by means of an offering memorandum and related subscription agreement. The material is intended only to facilitate your discussions with Fulcrum Asset Management as to the opportunities available to our clients. The given material is subject to change and, although based upon information which we consider reliable, it is not guaranteed as to accuracy or completeness and it should not be relied upon as such. The material is not intended to be used as a general guide to investing, or as a source of any specific investment recommendations, and makes no implied or express recommendations concerning the manner in which any clients account should or would be handled, as appropriate investment strategies depend upon clients investment objectives. Funds managed by Fulcrum Asset Management LLP are in general managed using quantitative models though, where this is the case, Fulcrum Asset Management LLP can and do make discretionary decisions on a frequent basis and reserves the right to do so at any point. Past performance is not a guide to future performance. Future returns are not guaranteed and a loss of principal may occur. Fulcrum Asset Management LLP is authorised and regulated by the Financial Conduct Authority of the United Kingdom (No: 23683) and incorporated as a Limited Liability Partnership in England and Wales (No: OC3641) with its registered office at Marble Arch House, 66 Seymour Street, London, W1H 5BT. Fulcrum Asset Management LP is a wholly owned subsidiary of Fulcrum Asset Management LLP incorporated in the State of Delaware, operating from 35 Park Avenue, 13th Floor New York, NY 122. c 218 Fulcrum Asset Management LLP. All rights reserved. 1