A Financial Conditions Index for Norway

Vonen, Nikka Husom Research Report A Financial Conditions Index for Norway Staff Memo, No. 07/2011 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Vonen, Nikka Husom (2011) : A Financial Conditions Index for Norway, Staff Memo, No. 07/2011, ISBN 978-82-7553-613-4, Norges Bank, Oslo, https://hdl.handle.net/11250/2507329 This Version is available at: https://hdl.handle.net/10419/210223 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/deed.no No. 07 | 2011 A financial conditions index for Norway Nikka Husom Vonen, Monetary Policy Staff Memo Staff Memos present reports and documentation written by staff members and affiliates of Norges Bank, the central bank of Norway. Views and conclusions expressed in Staff Memos should not be taken to represent the views of Norges Bank. © 2011 Norges Bank The text may be quoted or referred to, provided that due acknowledgement is given to source. Staff Memo inneholder utredninger og dokumentasjon skrevet av Norges Banks ansatte og andre forfattere tilknyttet Norges Bank. Synspunkter og konklusjoner i arbeidene er ikke nødvendigvis representative for Norges Banks. © 2011 Norges Bank Det kan siteres fra eller henvises til dette arbeid, gitt at forfatter og Norges Bank oppgis som kilde. ISSN 1504-2596 (online only) ISBN 978-82-7553- (online only) *The views in this paper are those of the author and should not be interpreted as those of Norges Bank. I am grateful to Karsten Gerdrup, Kjetil Olsen, Knut Are Aastveit and Steinar Holden for valuable help and comments on the work leading up to this paper. A financial conditions index for Norway* Nikka Husom Vonen Norges Bank July 2011 Abstract Financial conditions indexes (FCIs) may be useful tools for policymakers because they may have the ability to summarize overall financial conditions for households and companies and at the same time provide timely information on real economic activity. A monthly FCI for Norway is constructed by using principal components based on 13 financial variables. Real-financial linkages are examined by correlation analyses and by analyzing inand out-of-sample fit of a regression model. Through this exercise the FCI is found to be a useful leading indicator of Norwegian GDP growth. Alternative FCIs are also considered, in particular an FCI based on variables purged of business cycle effects. 2 1 Introduction The recent financial crisis and the economic downturn that followed have brought to the fore the importance of financial conditions and how they may affect the overall economy. As financial markets develop and come to play a greater role, the information set available to policymakers expands as new financial products emerge. At the same time, it becomes harder to gauge the overall state of financial conditions, with individual variables possibly giving mixed signals. This poses a challenge for monetary policy. In preparing for policy decisions, Norges Bank analyzes the state of the economy. This work is partly guided by models. The financial sector and financial factors are, however, far from adequately captured in models used for policy (Woodford, 2010). Furthermore, a number of indicators of financial conditions are observed and subject to discretionary judgment. As the information set is vast, it can be useful to follow summary statistics capturing a broader picture in a few single measures. A financial conditions index (FCI) may be suitable in this respect. Springing out from the literature on monetary conditions indexes (MCIs), intended to capture the overall stance of monetary policy, the more comprehensive FCIs are created to provide information about the broader financial conditions and their impact on economic activity. As methods and financial variables differ between FCIs, the exact focus, use and interpretation varies across indicators. In some cases FCIs measure the tightness/accommodativeness of financial factors relative to their historical average, while other indexes illustrate financial conditions’ contribution to growth. Some indexes are closely related to policy making in as much as index values can be interpreted in terms of interest rate equivalents (see e.g. Beaton, Lalonde and Luu (2009)). Other indexes are more oriented towards forecasting and may be used as leading indicators as they can provide timely information about economic activity. Whether an FCI mainly captures financial variables’ response to economic activity, or if it is more of an indicator of financial conditions’ impact on real activity, depends on how it is constructed, although this distinction is not always made clear. However, in both instances an FCI can provide early and leading information as financial data typically are available well in advance of quarterly national statistics. A number of FCIs have already been constructed, both by central banks, international organizations and private institutions. However, no such indicator has so far been developed for 3 the Norwegian economy. Inspired by existing FCIs for other economies and adapting to a Norwegian framework, the construction of a financial conditions index for Norway is thus the main focus of this paper. 1 , 2 Several methods have been used to construct FCIs. Here, an underlying factor is estimated by using the method of principal components. This procedure allows for the inclusion of a large number of financial variables, yet a parsimonious model is retained. The estimated factor is taken as a measure of financial conditions, which in turn are expected to provide information about real economic activity. To examine the possible realfinancial linkages, correlation analyses and analyses of inand out-of-sample fit of a regression model are performed and supplemented with graphical inspections. The FCI is found to be a useful leading indicator of real economic activity: It is easy to estimate and available on a monthly basis. An alternative version of the index is also considered, where the FCI is based on financial variables purged of business cycle effects. This adjustment is done in order to create an FCI reflecting financial variables’ impact on economic growth, without including their endogenous response to real economic activity. Although both versions of the index can be used, the unpurged version is chosen as the preferred FCI for Norway. The rest of the paper is organized as follows: Further motivation for and discussion of an FCI is provided in section 2, while the method is presented in section 3. Data and related issues are discussed in section 4. In section 5 principal components are estimated and real-financial linkages are examined through correlation analysis and by analyzing in-and out-of-sample fit of a regression model. In section 6 the preferred FCI for Norway is presented, together with an alternative FCI purged of business cycle effects. A few words of caution are noted in section 7, before a short conclusion is provided in section 8. 1 This paper provides a summary of the work presented in Vonen (2011). 2 Work related to the role of financial conditions in Norway does however exist, see e.g. Langbraaten (2001) and Bjørnland and Jacobsen (2010) on the role of asset prices for monetary policy, Gerdrup, Hammersland and Naug (2006), Aastveit and Trovik (2010) and Næs, Skjeltorp and Ødegaard (2011) on the relationship between financial variables and the real economy, and Brubakk and Natvik (2010) for the inclusion of financial frictions in a policy model. 4 2 The importance of financial conditions and the role of FCIs As monetary policy makes itself felt by altering financial conditions, an FCI can be expected to be useful for policymakers. Following Hatzius et al. (2010, p. 1), financial conditions can be defined as “(...) the current state of financial variables that influence economic behavior and (thereby) the future state of the economy.” Descriptions of the monetary transmission mechanism are thoroughly covered elsewhere; see e.g. Boivin, Kiley and Mishkin (2010) and Norges Bank (2004). To put it shortly, a wide range of variables may have a bearing on households’ and firms’ spending and investment behavior. In addition to interest and exchange rate effects, asset prices such as house and stock prices are of importance. Furthermore, credit availability, risk premiums and liquidity conditions in various financial markets are also relevant features of the overall financial conditions. Data for such measures are thus relevant components of an FCI. There are several reasons to believe that asset prices (and other financial variables) can provide valuable information about current and future economic activity (see Gerdrup, Hammersland and Naug (2006), among several others). First of all, asset prices reflect expectations about future economic developments, and thus they are often said to be forward-looking. Second, asset prices may affect the economy with a lag, and finally, asset prices are updated frequently with a rather short time lag, and are seldom subject to revisions. Financial variables are both affected by and have an impact on the real economy. What is less obvious is how to assess the quantitative effects and strengths of the various links. Complicating things further, there are also other factors than monetary policy that affect financial conditions. And finally, the links between the policy rate and financial conditions, as well as those between financial conditions and real activity, are likely to vary (Hatzius et al., 2010). In light of the possible interactions and effects mentioned above, the policymaker’s task of deciding on the appropriate level of the policy rate, or on the use of less conventional policy instruments for that matter, may seem daunting. A lot of information is subject to discretionary judgment. Individual variables may give mixed signals, and the appropriate weight given to various aspects is not necessarily known. This is where a financial conditions index (FCI) may be introduced as a helpful, albeit crude tool. A single indicator capturing the overall financial conditions may provide a useful guideline when making policy decisions. FCIs have already been constructed for several countries, based on 5 various methodologies and taking on different interpretations and uses. The literature on FCIs may be seen as springing out from the work on the more narrow monetary conditions indexes (MCIs) during the 1990s (Hatzius et al. (2010). These MCIs were weighted averages of an interest rate and an exchange rate, attempting to capture the overall stance of monetary policy. Over time, more variables have been included, and other methodological approaches have been employed in order to construct broader indicators, known as financial conditions indexes. The methods most commonly used can be divided in two broad groups: (i) A weighted-sum approach and a (ii) principal components (PC) approach (Hatzius et al., 2010). (i) In the first approach, each variable is given a weight in the FCI according to the relative impact of a change in that variable on real economic activity. These weights are in turn determined in one of three different ways: By estimating a structural/large scale macro model, by using reduced form demand equations or by use of vector autoregressions (VAR). (ii) In the second approach, one or several underlying factors are estimated from a larger set of variables. As these factors are unknown, they need to be estimated, and this is often done by use of the principal components method. The first such principal component captures the largest share of variation among the included variables, and this primary principal component is often used as an FCI in itself. For construction of the Norwegian FCI, the latter approach is chosen, and this method is briefly described in the following section. 3 Methodology – estimating factors by principal components A number of the most recently published papers on FCIs make use of factor estimation for FCI construction, using one or combinations of several estimated factors as an FCI. Relevant examples are Hatzius et al. (2010), Brave and Butters (2010) and Skaarup, Duschek-Hansen and Nielsen (2010), among others. One attractive feature of this approach is that it allows for the incorporation of a broad number of variables in a parsimonious way. This contrasts the weightedsum approach mentioned above, where only a small number of varibles can be included. The approach taken in paper is closely related to the work of Hatzius et al. (2010) and Skaarup, Duschek-Hansen and Nielsen (2010). For a more in-depth and technical derivation of factor 6 models and the use of principal components, see e.g. Stock and Watson (2002a); here, a brief description is provided. 3 Many macroeconomic time series tend to co-vary over time, and this has been exploited in business cycle analysis for several decades (Stock and Watson, 2002b). One may hypothesize that a few underlying factors govern the movement in a larger number of series. Assuming such a factor representation of the data is appropriate, factors are related to the observable variables (in this case a set of financial variables) in the following way: Let N be the number of variables , and T be the number of time period observations included in the analysis, . The time t observation of a given variable can then be expressed as (3.1) where is the underlying factor. The relationship between a given factor and an observable variable is given by the so called factor loadings, .These loadings will in general differ between the variables, and for each variable there is one factor loading associated with each of the underlying factors. is referred to as the common component of the model. is the idiosyncratic or variable specific component reflecting the “uniqueness” in each variable, that is, the part of the variation in a series which is not common to all the included variables. The underlying factors are not observable themselves, and therefore they need to be estimated. As for several other FCIs, principal components are used for this purpose. The first principal component accounts for the largest share of total variance in the data. The next principal components are labeled according to the declining share of variance accounted for. Note that all the principal components are orthogonal to each other; a given principal component is uncorrelated with all the other principal components. In total, the number of principal components is equal to the number of original variables in the dataset. 4 However, a substantial 3 The following paragraphs are largely based on Stock and Watson (2002a), Johnson and Wichern (1992) and Theil (1971). 4 In order to make the variables comparable, they are standardized before being transformed to principal components. Standardization implies that the variance of each variable equals one, and therefore the total variance in the dataset is equal to the number of variables . 13 Figure 5.3: Factor loadings, first principal component. Most of the variables have positive factor loadings, while the two illiquidity measures have negative loadings. As the factor representation in itself gives no information about the relationship between the financial variables and economic growth, it may be hard to give an economic interpretation of the factor estimates and the corresponding loadings. However, anticipating the course of events, the regression results reported below reveal a significant positive relationship between the first principal component and quarterly growth in GDP. This finding will facilitate the interpretation of the factor loadings. If a given factor loading is positive, an increase in the corresponding variable is positively associated with an increase in the underlying factor, this factor being represented by the first principal component. As this principal component is positively related to growth, one can argue in favour of a positive relationship between the corresponding variable and growth.  The magnitude of the factor loadings varies substantially between the variables, but qualitatively, the signs of the loadings are possible to reconcile with the expected relationship between the financial variables and GDP growth. For example, stock and house prices are likely to be positively related to growth, even though causation can go in both directions. A depreciated currency may stimulate growth through an increase in net exports, and money and credit measures are also expected to be positively related to growth, even though the question of timing may be a relevant issue. The positive loading on the NIBOR rate may be more surprising. Higher interest rates are often seen to have a 0,20 0,08 0,38 0,32 0,25 0,30 0,14 0,36 0,35 0,26 -0,42 -0,18 0,07 -0,6 -0,4 -0,2 0 0,2 0,4 0,6 14 dampening effect on growth. On the other hand, stronger growth calls for higher interest rates, and hence the association between growth rates and interest rates can be both positive and negative. Unfortunately, the method does not allow for a detailed interpretation of exactly which effects that are captured by the principal components, and causation cannot be established.  The two illiquidity measures have negative loadings. This is in line with what is expected, as stock market illiquidity may be an indication of a future economic downturn. Such empirical findings and possible explanations are described in detail in Næs, Skjeltorp and Ødegaard (2011). As has been mentioned already, causal effects can go in several directions and exactly which mechanisms that are captured by the principal component(s) and what effects are reflected in the loadings is not clear. However, the considerations behind the variables chosen, the signs of the estimated factor loadings and the regression results reported below indicate a positive link between the first principal component and real economic activity. Therefore, this first principal component may provide a useful basis for an FCI. Before a more formal investigation of the relationship between the first principal component and growth is carried out, some graphical plots are provided. As mentioned above, quarterly series of the principal components are needed for comparison with GDP figures. When every third observation is used, the following quarterly series for the first principal component is obtained: Figure 5.4: First principal component (PC1Q), quarterly series. 1994Q1-2010Q4. -8 -6 -4 -2 0 2 4 6 -8 -6 -4 -2 0 2 4 6 1994 1996 1998 2000 2002 2004 2006 2008 2010 15 Naturally, this quarterly series is less smooth than the original monthly series, but the overall picture is retained. In figure 5.5a the quarterly principal component is plotted together with quarterly growth in seasonally adjusted GDP. The two series do not track each other particularly well. However, the latter series is very volatile and development in this series is thus rather hard to predict. Figure 5.5a and b: First quarterly principal component (PCQ1) and a) quarterly GDP growth, sa (%) b) four-quarter GDP growth, sa (%).1994Q1-2010Q4 However, the PCQ1 and four-quarter growth in the same GDP series follow a more similar pattern, as seen in panel b of figure 5.5. This may indicate that the first principal component picks up a more cyclical pattern of lower frequency. This issue will be further examined in section 6.1. So far the discussion has been concentrated on the first principal component (and its quarterly counterpart). The next principal components, more specifically PC2-PC4, exhibit quite different patterns compared to PC1. This is only to be expected as principal components are orthogonal to each other by construction. However, as they may carry relevant information about real economic activity, these PCs are included in the regressions described below. 5.2 The predictive power of the financial factor(s) – how close is the link between financial conditions and real economic activity? In order to assess the link between financial conditions – as captured by the principal components - and GDP growth more formally, both in-sample and pseudo out-of sample regressions are performed. Quarterly growth in seasonally adjusted GDP is regressed on one lag of itself and on one or more of the quarterly principal components, which are also lagged relative to the quarterly -8 -6 -4 -2 0 2 4 6 -8 -6 -4 -2 0 2 4 6 1994 1996 1998 2000 2002 2004 2006 2008 2010 PC1Q GDP sa (q/q) % -4 -2 0 2 4 6 8 -8 -6 -4 -2 0 2 4 6 1994 1996 1998 2000 2002 2004 2006 2008 2010 FCIQ GDP sa, four-quarter growth, % (rhs) 16 observation being explained. A constant term is also included in all the regressions. The general model being estimated is given by the following equation ∑ (j=1,…4, h=1,…5) (5.1) where is quarterly GDP growth, is the time t observation of principal component j and h is the quarterly forecast horizon. As seen from table 5.1 above, the first four principal components account for the bulk of the variation in the dataset. Therefore, these four PCs are used as explanatory variables by including one, two, three or four principal components. 12 Regressions are first done for the whole sample period (in-sample), after which a recursive pseudo out-ofsample forecast is done. Results are initially compared to those of a simple autoregressive model. At a later stage, results will also be compared to those from regressions using individual financial variables, see section 5.2.3. In the remaining part of this chapter, regression results are provided and briefly commented on. 5.2.1 In-sample regressions using principal components R2 and p-values for the overall regressions are reported in table 5.2. 13 The autoregressive model, where GDP is regressed on one lag of itself and a constant term hardly has any explanatory power and is not significant. 14 This is not surprising, bearing in mind the high volatility of quarterly GDP growth in Norway, and that only one lag of the dependent variable is considered. 15 By including principal component(s) in the model, the results are significantly improved, with the best result found for the model where only the first principal component is included, predicting growth one quarter ahead (h=1). Admittedly, R2 is still not very large, but all models are significant at a 5 % level, both for one and two quarters ahead. 12 Either the first, the two first, the three first or the four first. Results are only provided for h=1 and h=2. 13 The rows of the table refer to different models: E.g. “3 PC” indicates the model including the three first principal components. This pattern is followed throughout the paper. 14 Here “significant” refers to p-values below 0.05. 15 One lag is used for all the regressions (except when purging the financial variables in section 6.2, see equation 6.1). A higher number of lags could have been included, either a fixed number or determined by some information criterion. 17 Table 5.2: In-sample regression results for AR and PC models. h=1 and h=2. Horizon: h=1 R² p-value AR 0.0185 0.2726 1 PC 0.2447 0.0001 2 PC 0.2460 0.0005 3 PC 0.2460 0.0014 4 PC 0.2864 0.0008 Horizon: h=2 R² p-value AR 0.0516 0.0665 1 PC 0.1540 0.0051 2 PC 0.1899 0.0043 3 PC 0.2065 0.0063 4 PC 0.2102 0.0127 5.2.2 Pseudo out-of-sample forecasting Next, pseudo out-of-sample forecasting is performed.The models are estimated recursively, with forecasts starting in 2001Q1. The forecast period is further split into two sub-periods: 2001Q12005Q4 and 2006Q1-2010Q4. The models are evaluated in terms of root mean square errors (RMSEs) and relative RMSEs with the simple AR-model as a benchmark. Forecast results are reported in table 5.3. Table 5.3: Pseudo out-of-sample forecast results for AR and PC models, h=1 and h=2. The lowest relative RMSEs for the whole forecast period are highlighted. h=1 RMSE Relative RMSE Period 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 AR 0.9461 0.8356 1.0451 1 1 1 1 PC 0.7985 0.7511 0.8432 0.8440 0.8989 0.8068 2 PC 0.8128 0.7608 0.8616 0.8591 0.9105 0.8244 3 PC 0.8846 0.8882 0.8809 0.9350 1.0629 0.8429 4 PC 0.8666 0.8651 0.8652 0.9160 1.0353 0.8279 h=2 RMSE Relative RMSE Period 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 AR 0.8920 0.8210 0.9577 1 1 1 1 PC 0.8507 0.7738 0.9212 0.9537 0.9425 0.9619 2 PC 0.8556 0.7870 0.9190 0.9592 0.9586 0.9596 3 PC 0.8942 0.8842 0.9041 1.0025 1.0770 0.9440 4 PC 0.9343 0.9358 0.9329 1.0474 1.1398 0.9741 18 The best forecast result in terms of relative RMSE for the whole forecast period is found for the model including the first principal component. This is true both for one and two quarters ahead, even though results for the former is better. There are some differences between the two subperiods. For one quarter ahead, the absolute RMSEs are quite similar across the two sub-periods. The change in relative RMSE can thus be traced back to the particular poor AR forecast for one period ahead. Results for the whole forecast period indicate that the models with either the first or the two first principal components are more relevant for predicting GDP growth. 5.2.3 Comparison with individual financial series An autoregressive model with only one lag may seem like a somewhat sparse basis for comparison. Therefore, similar regressions are performed where the principal components are replaced with individual financial variables. More specifically, the in equation 5.1 is now taken to represent a single financial variable rather than one of the estimated PCs. Models are estimated for all the 13 financial variables listed in table 4.1. As noted above, attempts at predicting real activity using financial data abound. Since the principal components are derived from the same financial series, the relevant question here is whether there is something to gain from pooling the information compared to the performance of forecasts based on individual financial series. There is reason to believe, and empirical evidence shows that this may indeed be the case, as idiosyncratic variation in the included series in a sense is averaged away (Stock and Watson, 2002b). Inand out-of-sample results are reported in tables 5.4a and b, with the AR and PC1 results included for ease of comparison. For one quarter ahead, the “best” individual financial variable is the relative spread (RS), with insample results similar to those of the PC1 model. The relative spread is also the best single series out-of-sample for the whole forecast period taken together. However, the PC1 model has a lower relative RMSE than any of forecasts based on individual financial series. In total, five variables are significant explanatory variables in-sample (highlighted in the table), and these are also among the best-performing variables out-of-sample. For two quarters ahead, table 5.4b reveals that house prices is the best single series, marginally outperforming the PC1 model both inand out-of-sample. M1 yields results similar to the PC1 model, but for the other individual series results are somewhat poorer. 19 Table 5.4a: Inand out-of-sample results for individual financial variables, h=1. In-sample Out-of-sample – different periods h =1 RMSE Relative RMSE R² p-value 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 10yr3mth 0.1572 0.0042 0.8857 0.7326 1.0160 0.9362 0.8767 0.9722 TWI 0.0199 0.5259 1.0860 1.1175 1.0535 1.1479 1.3374 1.0080 Stock 0.1657 0.0030 0.8834 0.7221 1.0195 0.9337 0.8642 0.9755 House 0.0791 0.0715 0.9146 0.8434 0.9806 0.9667 1.0093 0.9383 Oil 0.0335 0.3356 0.9603 0.8516 1.0578 1.0150 1.0191 1.0122 C1 gp 0.0583 0.1465 0.9389 0.8440 1.0250 0.9924 1.0101 0.9808 C2 nfe 0.0209 0.5080 0.9570 0.8430 1.0587 1.0115 1.0089 1.0130 C2 bank 0.1157 0.0195 0.9199 0.8398 0.9936 0.9723 1.0050 0.9507 M1 0.1145 0.0204 0.8812 0.8158 0.9420 0.9314 0.9763 0.9013 M2 nfe 0.0579 0.1484 0.9340 0.8537 1.0079 0.9872 1.0217 0.9644 RS 0.2172 0.0004 0.8667 0.7786 0.9466 0.9161 0.9318 0.9058 Amihud 0.0190 0.5411 0.9741 0.8243 1.1040 1.0296 0.9865 1.0564 NIBOR 0.0189 0.5431 0.9707 0.8546 1.0744 1.0260 1.0227 1.0280 AR 0.0185 0.2726 0.9461 0.8356 1.0451 1 1 1 PC1 0.2447 0.0001 0.7985 0.7511 0.8432 0.8440 0.8989 0.8068 Table 5.4b: Inand out-of-sample results for individual financial variables, h=2. In-sample Out-of-sample – different periods h =2 RMSE Relative RMSE R² p-value 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 2001Q12010Q4 2001Q12005Q4 2006Q12010Q4 10yr3mth 0.1090 0.0264 0.8561 0.7889 0.9184 0.9598 0.9609 0.9590 TWI 0.0560 0.1630 0.9554 0.9442 0.9665 1.0711 1.1501 1.0092 Stock 0.0895 0.0521 0.8797 0.7995 0.9531 0.9862 0.9738 0.9952 House 0.1869 0.0015 0.8424 0.8391 0.8456 0.9444 1.0220 0.8829 Oil 0.0628 0.1297 0.9024 0.8052 0.9901 1.0117 0.9808 1.0338 C1 gp 0.0517 0.1880 0.9083 0.8414 0.9706 1.0183 1.0248 1.0135 C2 nfe 0.0517 0.1880 0.9044 0.8247 0.9777 1.0139 1.0045 1.0209 C2 bank 0.0625 0.1311 0.9009 0.8248 0.9710 1.0100 1.0046 1.0139 M1 0.1426 0.0079 0.8556 0.8313 0.8792 0.9592 1.0125 0.9180 M2 nfe 0.0725 0.0935 0.8955 0.8392 0.9484 1.0039 1.0222 0.9903 RS 0.1017 0.0341 0.8778 0.8012 0.9482 0.9841 0.9759 0.9901 Amihud 0.1030 0.0326 0.8979 0.8118 0.9764 1.0066 0.9888 1.0195 NIBOR 0.1036 0.0319 0.9297 0.8325 1.0177 1.0423 1.0140 1.0627 AR 0.0516 0.0665 0.8920 0.8210 0.9577 1 1 1 PC1 0.1540 0.0051 0.8507 0.7738 0.9212 0.9537 0.9425 0.9619 20 Summing up, the best individual financial variables yield similar results to those from models including one or several principal components, although which variable is better depends on the forecast horizon. Somewhat unstable results across forecast periods are also present for the forecasts based on individual financial series, and this problem is therefore not just confined to the models based on principal components. However, no formal in-sample test for structural instability is made, and it is therefore hard to do any further formal comparison between the PC models and models including individual financial series in this respect. However, combining the inand out-of-sample results, the overall impression is that there seems to be something to gain from pooling information by using estimated factors. Even though individual variables occasionally perform better, one could claim that it is more robust to make use of the estimated factors as these yield better results “on average.” 6 An attempt at an FCI for Norway The analysis described above indicates a significant positive relationship between the first principal component and GDP growth. Furthermore, there may be useful information in some of the other principal components as well, and so it remains to be decided exactly how the FCI should be constructed. Most FCIs based on estimated factors only include the first estimated factor/principal component. However, in some studies FCIs are constructed by weighing together several principal components, such as in Gomez, Murcia and Zamudio (2011) and English, Tsatsaronis and Zoli (2005). However, as most of the relevant information for Norwegian GDP growth seems to be captured by the first principal component, this series alone is taken to be the FCI for Norway. This also facilitates the interpretation of the index, as indicated in the discussion of the factor loadings. Moreover, it is conceivable that the other principal components pick up rather different developments as these are orthogonal to the first principal component. The time series plotted in figure 5.1 is thus taken to be the preferred FCI for Norway. It is easy to calculate and it can be updated every month. There is a positive relation between the FCI and GDP growth, hinting at its possible use as a leading indicator of real economic activity. However, the indicator probably captures both impacts from financial factors to the real economy, as well as the financial variables’ endogenous response to economic activity. Further comments on this distinction is made in section 6.2 where an alternative index is presented. 21 6.1 Use and interpretation of the FCI The preferred FCI for Norway provides a summary measure of the broad financial conditions, and can further be indicative of developments in the real economy. However, as already mentioned, quarterly growth in Norwegian GDP is particularly volatile and therefore hard to predict. In a final attempt to establish a link between the FCI and real economic activity, the FCI is related to alternative measures of real activity. In particular, the quarterly FCI (as seen in figure 5.4) is plotted together with growth (both quarterly and four-quarter growth) in trend GDP, a smoother time series. 16 In general, the trend version of a series may be more informative than the seasonally adjusted version if there is a lot of random variation in the raw data. In addition to graphical plots, as seen in figures 6.1 and 6.2, correlation coefficients between the FCIQ and the trend growth series are calculated, see table 6.1. This is done both for the contemporaneous relationship between the variables, as well as including leads, where the time t value of the FCIQ is related to growth in future quarters (t+1, t+2 etc). Figure 6.1: Quarterly FCI and quarterly growth in trend GDP (%). 1994Q1-2010Q4. 16 Time series are often seasonally adjusted to remove variations in the series related to seasonal effects – fluctuations in the series that typically occur around the same time every year. One way of adjusting a time series it to decompose the series into three parts (after a pre-treatment of the data): A seasonal component (S), a trend component (T) and an irregular component (I). The seasonal component contains the variation in the time series that occurs within one year, and a seasonally adjusted series (A) thus only contains the trend and irregular components (A=T+I). The trend component is a smoother series reflecting a more underlying tendency in the data. This measure may be more informative if the original series contains a lot of random/ irregular variation. For further details, see http://www.ssb.no/english/metadata/methods/seasonal_adjustment.pdf. -1 -0,5 0 0,5 1 1,5 2 -8 -6 -4 -2 0 2 4 6 1994 1996 1998 2000 2002 2004 2006 2008 2010 FCIQ GDP trend (q/q), % (rhs) 22 Figure 6.2: Quarterly FCI and four-quarter growth in trend GDP (%). 1994Q1-2010Q4. Table 6.1: Correlations between quarterly FCI and trend growth in GDP (the highest value for each variable is highlighted). Correlations with FCIQ t t+1 t+2 t+3 t+4 t+5 t+6 GDP trend (q/q) 0.69 0.65 0.50 0.35 GDP trend (y/y) 0.64 0.76 0.78 0.71 Both the graphs and correlation coefficients lend support to the interpretation of the FCI being a leading indicator of real economic activity. 17 Admittedly, for quarterly trend growth, the FCI is more of a coincident than a leading indicator, as the contemporaneous correlation coefficient is the highest one. Still, it would in practice be leading as the financial data on which it is based are available well in advance of actual GDP data. On the other hand, the FCI is clearly leading developments in four-quarter trend growth, with the highest correlation coefficient found for two quarters ahead. A correlation of around 0.8 indicates a rather close relationship between the FCI and trend growth in GDP. This is quite remarkable bearing in mind that the FCI is estimated on financial variables only. Some studies using principal components make an attempt at identifying or characterizing the estimated factors. One example is seen in Martinsen (2010, p. 19), where the results indicate that “(...) the first factors for the regions pick up the developments in real activity, 17 When relating the FCI to quarterly measures of real activity, the quarterly FCI, FCIQ, is used. However, the results are taken to be representative for the monthly FCI as well. -3 -1 1 3 5 7 -8 -6 -4 -2 0 2 4 6 1994 1996 1998 2000 2002 2004 2006 2008 2010 FCIQ GDP trend, four-quarter growth, % (rhs) 29 could also be seen as more intuitive than e.g. an interpretation in terms of standard deviations (Hakkio and Keeton, 2009). Finally, in interpreting the index the inherent lack of structural foundation of FCIs based on the principal components method should be kept in mind. Movements in the index can be traced back to changes in the underlying financial variables. However, even though attempts have been made to interpret the FCI and to establish correlation and to some extent causation, neither the exact economic mechanisms nor a fine-tuned picture can be revealed. This is not to undermine the potential usefulness and benefit of the FCI, but rather to emphasize that its value should be judged based on what it really is – a “rough and ready” summary statistic. 8 Summing up The work on a financial conditions index for Norway is summarized in this paper, while a broader coverage is provided in Vonen (2011). The first principal component from a dataset of 13 monthly financial variables is used as the preferred FCI for Norway. Even though the lack of structural foundation may complicate interpretations, the index is found to carry information about real economic activity. Two different versions of the index are suggested: One based on financial variables purged of the impact of GDP growth, while the other one is calculated without such adjustments. Even though the latter is given the most attention and is chosen as the preferred FCI, both versions can in principle be put to use, as long as the distinction between them is made clear: While the purged version may be interpreted as financial conditions’ impact on the real economy, the unpurged version also contains financial variables’ response to real economic activity. However, for the purpose of a leading indicator, the unpurged version is clearly a better measure. The real-financial linkages are examined both in-and out-of-sample. Even if quarterly growth in seasonally adjusted GDP is hard to predict, the FCI generally outperforms the alternative models examined. Furthermore, the link appears to be stronger when relating the FCI to slower-moving measures such as trend growth in GDP, especially four-quarter trend growth. Taken together, the FCI may be useful for policy making in providing a comprehensive measure of financial conditions and at the same time being a leading indicator of real economic activity. Furthermore, the FCI is easy to estimate and can be updated every month. Several changes and improvements 30 are possible and should definitely be pursued. However, the current FCI is seen as a useful initial step and can easily be put to use as a readily available summary statistic and leading indicator. 31 References Aastveit, K. A. and T. Trovik (2010): Nowcasting Norwegian GDP: The role of asset prices in a small open economy. Empirical Economics. Published in “Online FirstTM” Dec 15th 2010, available at http://www.springerlink.com/content/166854l582861615/ [13.04.2011] Beaton, K., R. Lalonde and C. Luu (2009): A financial conditions index for the United States. Discussion Paper 2009-11, Bank of Canada. Bjørnland, H. C. and D. H. Jacobsen (2010): The role of house prices in the monetary policy transmission mechanism in small open economies. Journal of Financial Stability, 6(4), 218-229. Boivin, J., M. T. Kiley and F. S. Mishkin (2010): How has the monetary transmission mechanism evolved over time? In: B. M. Friedman and M. Woodford (eds): Handbook of monetary economics vol. 3 (pp. 369-422), Amsterdam: Elsevier B.V. Available at http://www.sciencedirect.com/science/handbooks/15734498 [23.04.2011] Boivin, J. and S. Ng (2006): Are more data always better for factor analysis? Journal of Econometrics, 132(1), 169-194. Brave, S. and R. A. Butters (2010): Gathering insights on the forest from the trees: A new metric for financial conditions. Working Paper 2010-07, Federal Reserve Bank of Chicago. Brubakk, L. and G. Natvik (2010): Credit, asset prices and monetary policy. Staff Memo 13/2010, Norges Bank. English, W., K. Tsatsaronis and E. Zoli (2005): Assessing the predictive power of measures of financial conditions for macroeconomic variables. Bank for International Settlements Papers No. 22, 228-252. Gauthier, C., C. Graham and Y. Liu (2004): Financial conditions indexes for Canada. Working Paper 2004-22, Bank of Canada. Gerdrup, K., R. Hammersland and B. Naug (2006): Financial variables and developments in the real economy. Norges Bank Economic Bulletin 77(3), 133-146. Gomez, E., A. Murcia and N. Zamudio (2011): Financial conditions index: Early and leading indicator for Colombia? Unpublished. Available at http://www.banrep.gov.co/documentos/publicaciones/report_estab_finan/2011/Tema1_ma rzo2011.pdf [19.04.2011] Guichard, S., D. Haugh and D. Turner (2009): Quantifying the effect of financial conditions in the euro area, Japan, United Kingdom and United States. OECD Economics Department Working Papers No. 677, OECD Publishing. Hakkio, C. S. and W. R. Keeton (2009): Financial stress: What is it, how can it be measured, and why does it matter? Federal Reserve Bank of Kansas City Economic Review, (2), 5-55. Hatzius, J., P. Hooper, F. Mishkin., K. L. Schoenholtz and M. W. Watson (2010): Financial conditions indexes: A fresh look after the financial crisis. Paper prepared for the Annual 32 U.S. Monetary Policy Forum, University of Chicago School of Business, February 2010. Available at http://research.chicagobooth.edu/igm/events/conferences/2010usmonetaryforum.aspx [12.01.2011] Johnson, R. A. and D. W. Wichern (1992): Applied multivariate statistical analysis. Englewood Cliffs, N.J.: Prentice-Hall International, Inc. Langbraaten, N. (2001): Formuespriser – konsekvenser for pengepolitikken? Norges Bank Economic Bulletin, (4), 198-205. (Available in Norwegian only). Martinsen, K. (2010): Forecasting in the regional network – a factor model approach. Master’s thesis, University College London. Murray, J. (2009): When the unconventional becomes conventional – monetary policy in extraordinary times. Speech by the Deputy Governor of the Bank of Canada. Bank for International Settlements Review 61/2009. Available at http://www.bis.org/review/r090522d.pdf?frames=0 [18.03.2011] Norges Bank (2004): Norske finansmarkeder – pengepolitikk og finansiell stabilitet. Norges Bank Occational Papers No. 34 (Available in Norwegian only). Norges Bank (2009): Monetary Policy Report 1/09. Reports from the Central Bank of Norway No. 1-2009. Norwegian Official Report (2011): Bedre rustet mot finanskriser. Finanskriseutvalgets utredning. The Ministry of Finance. Available at http://www.regjeringen.no/pages/15296179/PDFS/NOU201120110001000DDDPDFS.pdf [09.02.2011] (Available in Norwegian only). Næs, R., J. Skjeltorp and B. A. Ødegaard (2011): Stock market liquidity and the business cycle. The Journal of Finance, 66(1), 139-176. Skaarup, M., C. Duschek-Hansen and S. Nielsen (2010): A financial conditions index for Denmark. Working Paper no 23/2010, The Danish Ministry of Finance. Available at http://www.fm.dk/Publikationer/Arbejdspapirer/2010/~/media/Publikationer/arbejdspapir er/2010/arbejdspapir_financial%20conditions%20index%20for%20denmark.ashx [30.03.2011] Stock, J. H. and M. W. Watson (2002a): Forecasting using principal components from a large number of predictors. Journal of the American Statistical Association, 97, No. 460, 11671179. Stock, J. H. and M. W. Watson (2002b): Macroeconomic forecasting using diffusion indexes. Journal of Business and Economic Statistics, 20(2), 147-162. Stock, J. H. and M. W. Watson (2003): Forecasting output and inflation: The role of asset prices. Journal of Economic Literature, 41(3), 788-829. Swiston, A. (2008): A U.S. financial conditions index. Putting credit where credit is due. Working Paper 08/161, International Monetary Fund. 33 Theil, H. (1971): Principles of econometrics. Amsterdam: North-Holland Publishing Company. Vonen, N. H. (2011): A financial conditions index for Norway. Master’s thesis, University of Oslo. Woodford, M. (2010): Financial intermediation and macroeconomic analysis. Journal of Economic Perspectives, 24(4), 21-44. 34 Data documentation The following transformations are used: First difference = 2, differences in logarithms =5. The remaining series enter in levels =1. Table A.1: Variables included in the FCI. Series number Description Short name Source Transformation 1 Spread between the interest rates on 10 year government bonds and three month government bills 10yr3mth Thomson Reuters 1 2 Trade weighted real exchange rate (TWI). Real exchange rate between NOK and the currencies of 25 trading partners TWI Norges Bank 1 3 Stock market index for the Oslo Stock Exchange. Total return, total market. Stock Thomson Reuters 5 4 House prices, seasonally adjusted. Price per square meter. House NEF, NFF, Finn.no and EconPöyry. The series is extended with a data series from the RIMINI database (RIMINI is a macro model formerly used in Norges Bank). 5 5 Oil price (Brent Blend), USD/barrel. Oil Thomson Reuters 5 6 Credit (C1) to the general public. Domestic credit in NOK. C1 gp Statistics Norway 5 7 Credit (C2) to non-financial enterprises. Domestic credit in NOK and foreign currency. C2 nfe Statistics Norway 5 8 Credit (C2) from commercial banks. Domestic credit in NOK and foreign currency provided by commercial banks. C2 bank Statistics Norway 5 9 M1. Narrow money stock. M1 Statistics Norway 5 10 M2 non-financial enterprises. Broader money stock. M2 nfe Statistics Norway 5 11 Relative spread. Stock market illiquidity measure RS Norges Bank 1 12 Amihud’s stock market illiquidity measure Amihud Norges Bank 5 13 Three month NIBOR. NIBOR Norges Bank 2 35 Table A.2: Additional data series Description Source GDP Mainland Norway, seasonally adjusted Statistics Norway GDP Mainland Norway, trend component Statistics Norway Key policy rate Norges Bank The Business tendency survey – confidence indicator. Unadjusted. Statistics Norway