Assessing and Combining Financial Conditions Indexes

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Assessing and Combining Financial Conditions Indexes Sirio Aramonte, Samuel Rosen, and John W. Schindler 2013-39 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Assessing and Combining Financial Conditions Indexes Sirio Aramonte, Samuel Rosen, and John W. Schindler∗ May 30, 2013 Abstract We evaluate the short horizon predictive ability of financial conditions indexes for stock returns and macroeconomic variables. We find reliable predictability only when the sample includes the 2008 financial crisis, and we argue that this result is driven by tailoringtheindexestothecrisisandbynon-synchronoustrading. Financialconditions indexes are based on a variety of constituent variables and aggregation methods, and wediscussasimpleprocedureforconsolidatingthegrowingnumberofdifferentindexes into a single proxy for financial conditions. Keywords: Financial conditions indexes, stock returns predictability, forecasting. JEL Classification: E32, G01, G17. ∗Board of Governors of the Federal Reserve System, Office of Financial Stability Policy and Research. Corresponding author email and phone number: sirio.aramonte@frb.gov, +1-202-912-4301. We would like to thank Katherine Boiles for research assistance, and participants to several Federal Reserve Quantitative SurveillanceGroupmeetings. WewouldalsoliketothankMalcolmSpittlerandTroyMathesonforproviding data and guidance for, respectively, the Citi Financial Conditions Index and the International Monetary Fund U.S. Financial Conditions Index, and Mark Carlson, Kurt Lewis, and William Nelson, and Roberto Cardarelli, Selim Elekdag, and Subir Lall for providing data and guidance for, respectively, their Financial MarketStressIndexandtheInternationalMonetaryFundU.S.FinancialStressIndex. Thisarticlerepresents the views of the authors, and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or other members of its staff. 1

1 Introduction The severity and the economic impact of the 2008 financial crisis have led to a proliferation of indexes that proxy for financial conditions or financial stress, which we collectively refer to as financial conditions indexes (FCIs). In this paper we evaluate whether FCIs can predict stock returns or innovations to macroeconomic variables over a one-month or a one-quarter horizon. We focus on predictability over relatively short horizons in order to capture the effect that the health of the financial system has on the broader economy, rather than longer run predictability arising from business cycle fluctuations. In addition, we suggest a simple procedure for aggregating the various FCIs into a single proxy for financial conditions. We find that most FCIs can predict monthly and quarterly returns on the S&P 500 and on a portfolio of financial companies, and also innovations to a number of macroeconomic variables - but only if the period around the 2008 financial crisis is included (2007-2012). Such narrow predictability could be the result of threshold effects, in that financial conditions matter only after they deteriorate sufficiently.1 A second possibility is that the FCIs we consider have predictive power because they typically include variables that, like the TED spread, had unusually large movements during the 2008 crisis, and such variables were included in the FCIs precisely because of their pronounced fluctuations in 2007 and 2008. Figure 1 shows the times series of the TED spread from 1986 onwards, and it clearly highlights the uniquely high level that characterized the 2008 financial crisis. A third possibility is that some of the predictive power is the result of non-synchronous data: most FCIs, for instance, include VIX, an implied volatility index derived from options on the S&P 500. These options trade for 15 minutes after trading on the underlying index ends (4:15pm versus 4:00pm Eastern Time), with the consequence that VIX on day t contains information that will be reflected in stock prices only on day t+1 – a fact that can generate spurious 1 InKashyap, Lamont, andStein(1994), forinstance, financingconstraintsatthefirmlevelonlybecome binding in tight-money environments. Hubrich and Tetlow (2012) find that macroeconomic dynamics crucially depend on financial stress, with stress being “of negligible importance in ‘normal’ times, but of critical importance when the economy is in a high-stress [...] state.” (page 30). 2

predictability. FCIs are typically designed to measure whether broad financial conditions are loose or tight by historical standards (for instance, the Bloomberg and Chicago Fed indexes), or whether the financial system is experiencing historically unusual stress (e.g., the Carlson, Lewis, and Nelson (2012) index). The importance of a well-functioning financial system to the broad economy is highlighted by the results in Bernanke and Blinder (1992), Kashyap, Stein, and Wilcox (1993), Peek and Rosengren (1997), Kashyap, Lamont, and Stein (1994), and Paravisini (2008), who show that tight monetary policy, binding capital ratios, and bank financing constraints can reduce the supply of credit. The effect is stronger in the case of small banks with less liquid assets (Kashyap and Stein (2000)), and the impact is felt more directly by small firms (Gertler and Gilchrist (1994), Khwaja and Mian (2008)) and bankdependent borrowers (Chava and Purnanandam (2011)). A tight credit supply ultimately affects investment (Campello, Graham, and Harvey (2010)), inventories (Kashyap, Lamont, and Stein (1994)), and the broader economy (Bernanke (1983), Peek and Rosengren (1997), Peek and Rosengren (2000), Calomiris and Mason (2003)). The existing evidence is mixed on whether FCIs should be thought of as coincident indicators, or early warning indicators. While some studies suggest that FCIs can predict selected economic variables, the results are often unstable across sub-periods, which could be due to the FCIs including variables that have prominently characterized recent crises. In addition, the results may point to predictive power only at business-cycle frequency, which means that FCIs could be proxying for periodic changes in economic activity, rather than for the state of the financial system. As an example, Hatzius, Hooper, Mishkin, Schoenholtz, and Watson (2010) focus on predicting macroeconomic variables over one and two quarters, and they find that the FCIs they study seldom have better predictive power than stock market returns. The authors note on page 21 that: “the better performance during the most recent five years (relative to both the 3

average of the alternative FCIs and KC Fed’s index as representative of one of the better performing FCIs) may reflect selection bias in our choice of variables to include in the index: naturally, our selection was governed in part by an understanding of the types of financial variables that were used for monitoring andmeasuringtherecentfinancialcrisis. Inthissense,wedidnotseektomitigate observer bias.” Our predictability analysis relies on simple predictive regressions of a set of stock returnsorinnovationstomacroeconomicvariablesonlaggedFCIvalues. Wekeeptheindexes in levels, instead of using changes, because the authors uniformly emphasize that their FCIs are ordinal measures of financial conditions/stress. Relative to Hatzius, Hooper, Mishkin, Schoenholtz, and Watson (2010), who also evaluate the predictive power of several FCIs, we study predictability at a monthly as well as quarterly horizon, in order to minimize the risk that we find predictability arising from business cycle effects. Second, we study a larger number of FCIs and a larger number of financial and macroeconomic variables. Third, we explicitly discuss whether the predictability that arises in coincidence with the 2008 financial crisis is hard-wired in the FCIs, in the sense that the variables included in the FCIs may have been chosen on the basis of whether they experienced large fluctuations in 2007 and 2008. Finally, as described in the next paragraph, we assess the statistical significance of predicability with a methodology that is robust to biases generated by the high persistence that typically characterizes FCIs. As is clear from the time series plots in Figure 2, FCIs tend to be quite persistent. The high autocorrelation that characterizes the FCIs is also evident in the confidence intervals for the autoregressive roots shown in Table 2.2 We report confidence intervals, rather than point estimates, to highlight that, after accounting for statistical uncertainty, the FCIs plausibly have autoregressive roots very close to one. In 5 cases, we are actually unable to reject the 2 The procedure for calculating the confidence intervals in Table 2 is described in the online appendix to Campbell and Yogo (2006). 4

hypothesisthattheFCIshaveaunitroot. Thehighpersistenceoftheindexesisimportantto us because we use them as predictors, and relying on standard asymptotics when testing the null of no predictability in the presence of a highly persistent predictor can be inappropriate, if other conditions detailed in Section 2 are met. Specifically, it can lead to a rejection rate that is inconsistent with the nominal size of the test. We therefore use, where applicable, the local-to-unit asymptotics procedure of Campbell and Yogo (2006), which corrects for this rejection bias. In the second part of our analysis we discuss a simple two-step methodology for combining the different FCIs into a single proxy for financial conditions. The large number of FCIs is itself indicative of the uncertainty that surrounds the measurement of financial conditions, and aggregating the individual indexes can help average out model uncertainty and identify a cleaner proxy. We first identify the five indexes that best summarize the information provided by the remaining FCIs. As detailed in Section 3, we do so on the basis of the adjusted R2 from regressions of changes in the principal component of all indexes – except for index i – on changes in index i. While it is possible that a low R2 is the result of an index being a unique and more accurate measure of financial conditions, the broad nature and the overlap of the variables included in the FCIs (like Treasury yields, or the implied volatility index VIX) renders such a possibility unlikely. In the second step we form all combinations of the five indexes, and select the “best” combination on the basis of how well it summarizes the information in the remaining FCIs, again on the basis of the adjusted R2 from regressions that, in order to minimize overfitting concerns, we run on several subsamples. In Section 2 we evaluate the predictive power of a comprehensive selection of FCIs, using stock returns and macroeconomic quantities as dependent variables. In Section 3 we discuss the merits of combining a subset of FCIs into a composite FCI. Section 4 concludes. 5

2 Assessing the Predictive Ability of FCIs The set of 12 FCIs that we consider includes indexes that (1) focus on the United States, (2) are available at a monthly or higher frequency, and (3) have a sufficiently long history (see Table 1, which also reports data sources and lists the abbreviated names we use in the paper).3 The FCIs are largely based on financial market variables, including implied volatilities, Treasury yields, yield spreads, commercial paper rates, stock returns, and exchange rates (see Kliesen, Owyang, and Vermann (2012) for a detailed list of variables that underlie a range of the FCIs we study here). Some FCIs only include a relatively small set of variables, as in the case of the St. Louis Fed Financial Stress Index. Other indexes, such as the Chicago Fed Adjusted National Financial Conditions Index, contain over one hundred variables. The constituents are aggregated either with a principal component analysis or through a weighted sum. In the latter case, weights are typically assigned subjectively by the authors, although a few of the indexes use more sophisticated methods. The Cleveland Financial Stress Index (CFSI), for instance, calculates weights dynamically based on the relative dollar flow observed in the Federal Reserve Board’s Flow of Funds statistical release (Z.1).4 All the indexes are expressed in terms of z-scores, with the exception of the Financial Stress Index of Carlson, Lewis, and Nelson (2012), which is expressed in terms of probabilities. Table 2 reports summary statistics for the FCIs, together with their value at the end of September 2008. For most FCIs, the September 2008 value was in the worst 10% of the 1995- 2012 sample, with the exception of the Morgan Stanley index, which mainly deteriorated into October 2008. Pairwise correlations among the FCIs (see Table 3) range between 22% and 96%, with the majority being above 70%. 3 In order to increase the power of our test, we require that the FCIs cover the stressful episodes that characterized the late 1990s (the Asian and Russian financial crises, LTCM collapse). For this reason, we do not include the HSBC Financial Clog Index (Bloomberg ticker HSCLOG Index) or the Westpac U.S. Financial Stress Index (Bloomberg ticker WRAISTRS Index) whose series start in 2007 and 1998, respectively. 4 See Oet, Eiben, Bianco, Gramlich, and Ong (2011) for details. 6

We evaluate the predictive power of the 12 FCIs we study with a series of monthly and quarterly predictive regressions of the form: y = α+β ×FCI +(cid:15) t t−1 t The FCIs enter the predictive regressions in levels, instead of changes, because financial conditions are likely to have real effects when they are unusually tight or unusually loose, rather than when they move within their normal range. We do not control for the predictive power of other variables, like the variance risk premium (Bollerslev, Tauchen, and Zhou (2009)), because the results from regressions in which FCIs are the only predictor already suggest that they lack reliable predictive power at the horizons we focus on. TheFCIcoefficientisestimatedwithOLS,andweassessitsstatisticalsignificancewith either heteroskedasticity consistent standard errors, or with the local-to-unity asymptotics procedure of Campbell and Yogo (2006). Local-to-unity asymptotics is useful in evaluating the statistical significance of persistent predictors, because, in such cases, the standard t-test can give a rejection rate that is inconsistent with its nominal size. In practice, we assume that each of the FCIs follows an AR(1) process, and use localto-unity asymptotics unless the autoregressive root of the FCI is sufficiently distant from one, in a sense defined below, or unless there is no correlation between the innovations to the FCI’s autoregressive process and the innovations in a regression of the predicted variable on the FCI. Note that both a persistent predictor and a non-zero correlation are necessary for the standard OLS asymptotics to be inappropriate. Using the notation in Campbell and Yogo (2006), we rely on heteroskedasticity consistent standard errors if the DF-GLS statistic is less than -5 (a more negative DF-GLS statistic shifts the confidence interval ˆ for the autoregressive root of the predictor away from one), or if the parameter δ (which 7

measures the correlation between the innovations) is equal to 0.5 The dependent variables in our predictive regressions are (a) returns on a broad market index and on various industry portfolios, and (b) autoregressive residuals of several economic variables. We consider monthly and quarterly returns on the S&P 500 and on seven equally weighted industry portfolios: finance, construction, manufacturing, transportation, wholesale trade, retail trade, and services. The macroeconomic variables we consider measure the availability of credit (total consumer credit, and commercial and industrial loans), the state of the housing market (housing starts), and manufacturing activity (durable goods orders, industrial production, and total manufacturing inventory).6 We first present the results that focus on whether FCIs can predict stock returns in Tables 5 through 9. For each portfolio/FCI combination we report the coefficient on the FCI (β ) and the regression root mean squared error (RMSE). We show an asterisk next FCI to a coefficient when the coefficient is statistically significant, that is when the Campbell and Yogo (2006) 90% confidence interval does not include zero. Table 5 shows results for the 1995-2012 sample, and it indicates that 11 of the 12 FCIs can predict returns on the finance industry portfolio, that four can predict the S&P 500, and that the Morgan Stanley index can predict returns on the finance and three additional industry portfolios. There is noticeable dispersion in the RMSEs across industry portfolios for a given FCI, with the construction and services portfolios generally showing the largest RMSE and the finance the lowest, but the RMSEs are remarkably similar across FCIs for a given portfolio. The statistically significant coefficients have the expected negative sign, indicating that higher financial stress is followed by lower returns. One potential source of predictability is non-synchroneity across markets: many FCIs, 5 Whenneeded,welinearlyinterpolatethevaluesobtainedfromthelookuptablesintheonlineappendix to Campbell and Yogo (2006), which only provide a discrete set of values for δˆ and of the DF-GLS statistic. 6 Stock returns are from CRSP through Wharton Research Data Services. The macroeconomic data are from the FRED database of the St. Louis Federal Reserve. See Table 4 for summary statistics and additional details on the construction of the industry portfolios. 8

for instance, include the S&P 500 option implied volatility index VIX, which is based on options whose trading ends 15 minutes after the trading for stocks does - a fact that could cause spurious predictability (see Atchison, Butler, and Simonds (1987) for a discussion of the effects of non-synchronous trading on the autocorrelation of equity index returns). We explore this possibility by running predictive regressions on monthly industry returns that skip the first day of each month. Table 6 shows that now only 5 of the 12 FCIs have statisticallysignificantpredictivepowerforthe“Finance”portfolio, downfrom11inTable5, suggesting that non-synchroneity does play a role, but is not the sole driver of predictability. The severity of the 2008 financial crisis naturally raises the question of whether the predictive power of the FCIs mainly arises from the events that started in early 2007, or whether it is also present in the broader sample. In Table 7 we report the coefficients and RMSEs estimated over the 1995-2006 sample, and the results highlight that, essentially, there is no predictability left. Later in this section we discuss whether the lack of predictive power outside of the 2008 crisis is due to predictability only being there during periods of financial stress, or whether it is the result of the FCIs being tailored to the recent financial crisis. The conclusions that we can draw from monthly returns also carry over to quarterly returns, for which results are reported in Tables 8 and 9. The number of FCIs with predictive power for returns on the S&P 500 or on the financial industry portfolio in the 1995-2012 sample is 5. Similar to the monthly analysis, returns on the remaining industry portfolios are not predictable. Excluding the first day of each quarter when computing the returns (untabulated results) does not change the statistical significance of the coefficients, although restricting the sample to the pre-crisis period (1995-2006, Table 9) all but eliminates the predictability,withtheexceptionoftheMorganStanleyindex,whichcannowpredictreturns on 5 portfolios. Note that the coefficients for the Morgan Stanley index are positive, while they were negative at the monthly horizon in the full sample (Table 5). The positive sign is consistent with the possibility that, already at the quarterly horizon, the predictive power of 9

theMorganStanleyindexmaybeduetobusinesscyclepredictability–poorcurrentfinancial conditions imply that, over a medium/long horizon, economic and financial conditions are more likely to improve than to further deteriorate, and stock returns will likely be positive. We now discuss whether the FCIs are informative about future innovations to the macroeconomic variables we mentioned earlier in this section. In order to implement the Campbell and Yogo (2006) procedure, the only covariate we include in the predictive regressions is one of the FCIs, and we account for autocorrelation in macroeconomic variable changes by using the residuals from log-change autoregressions as the dependent variable, where the number of lags is chosen on the basis of the Schwarz Bayesian information criterion.7 The first set of results, reported in Table 10, focuses on one-month ahead predictability, with the sample running from 1995 to 2012.8 The Chicago Fed, St. Louis Fed, Kansas City Fed, and IMF FCI indexes predict all the variables. Most other indexes predict at least 3 of the 6 variables, with only IMF FSI having statistically insignificant coefficients throughout. When the sample excludes the 2008 financial crisis (Table 11), however, the evidence in favor of predictability is weak, with most indexes being able to predict only one macroeconomic variable - typically industrial production. Similar conclusions can be drawn when focusing on quarterly horizons, as shown in Tables 12 and 13 (only the Morgan Stanley index can predict more variables in the short sample than in the full sample). One possible reason why the predictability we find in the 1995-2012 sample is not robust to the exclusion of 2007-2012 is that it is subject to threshold effects, in that only 7 At the monthly frequency, we use two lags for total consumer credit, commercial and industrial loans, industrial production, and total manufacturing inventory, and one lag for durable goods orders and housing starts. At the quarterly frequency, we use two lags for total consumer credit and total manufacturing inventory, one lag for commercial and industrial loans and industrial production, and zero lags for durable goods orders and housing starts. 8 While we expect a negative relation between a worsening of current financial conditions and future economicactivity,somecoefficientsinTables3and4arepositive. ThereasonisthattheCampbelland Yogo (2006) procedure requires a negative correlation between current innovations to the dependent variable and the predictor, hence we sometimes need to multiply the FCI by -1. We systematically check that the sign of the estimated coefficient is as expected: positive if the FCI (multiplied by -1 as applicable) signals stress when low, negative otherwise. 10

poor financial conditions can predict future returns or macroeconomic conditions. In Figure 3 we show a scatter plot of innovations to log-changes of total inventory against the St. Louis FCI, where observations for which the index is above its 80th percentile are highlighted in color. Red triangles indicate an observation from between 2007 and 2012 and green squares indicate an observation from outside that range. The evidence in Figure 3 is not supportive of a threshold effect, in that the negative relationship between FCI and log-changes in inventories is only evident in the observations from the recent crisis. We argue that the “recent financial crisis effect” arises either (1) because predictability is only present when investors expect that large dislocations are approaching (for instance, correlations can change, and FCIs ultimately measure correlations); or (2) because the variables underlying the many FCIs constructed after the 2008 crisis were chosen based on their movements during the crisis.9 It is difficult to empirically assess the merit of (1) and (2) above, not least because only one major financial crisis is included in the sample. OuropinionisthattheempiricalevidenceinfavoroftheFCIshavingreliablepredictive power is weak, especially in light of the fact that the FCIs are built by combining public data for typically highly liquid financial instruments - hence they can hardly be characterized as containing privileged information. 3 Efficiently Combining FCIs In the previous section we established that the FCIs we study have weak predictive power for broad economic developments. We now turn to the question of how to consolidate the increasingly large number of indexes into a single proxy for financial conditions. The implicit assumption in searching for the “best” combination is that the FCIs we focus on 9 Figure 3 is representative of other index/variable combinations. The unreported figures are available upon request from the authors. 11

provide broadly similar information, and that no FCI stands out by virtue of measuring financial conditions with greatly superior accuracy. We believe such assumption is validated by the largely comparable performance of the different indexes in the predictability analysis discussed above, and by the fact that the FCIs mostly include variables that capture broad macroeconomic trends (for example, Treasury yields, or S&P 500 option implied volatility). We propose a simple methodology to combine the various FCIs into a single proxy for financial conditions, which entails calculating the first principal component of a subset of appropriately chosen indexes. The FCIs themselves are already an aggregation of underlying variables, often based on a principal component analysis. The procedure we describe below can be seen as a higher level consolidation that aggregates across different variable sets and methodologies, with the objective of smoothing out transitory fluctuations and extracting a more informative proxy for financial conditions. First, we sort the individual indexes on the basis of how well they capture the information contained in the remaining FCIs. We measure this ability to capture information using the adjusted R2 from regressions10 of changes in the first principal component of all indexes except for index i on changes in index i. Letting i denote the FCI of interest, with i = 1,...,12, and “fpc” the calculation of the first principal component: PC = fpc({FCI } ) −i j j(cid:54)=i ∆PC = γ +δ ×∆FCI +ε −i i t In the interest of robustness, we also run a second set of regressions where the dependent and independent variables are one-lag autoregressive residuals of, respectively, ∆PC and −i ∆FCI . i It is, of course, possible for these regressions to yield a low R2 if index i does not span the remaining FCIs because it is a radically better proxy for financial conditions. As noted 10 We use robust regressions (Hamilton (1991)), which reduce the influence of outliers. 12

above, the overlap and the encompassing nature of the variables that underlie the different indexesmakessuchapossibilityunlikely. Table14reportstheadjustedR2softheregressions described above, which we run on two different samples, 1995-2006 and 1995-2012. The rankings in the two samples are quite similar, with the St. Louis FCI, in particular, having a noticeable margin on the other indexes. We use the ranking to select the five FCIs with the highest adjusted R2 for further aggregation. The two Bloomberg FCIs are ranked among the top five when the sample includes the period surrounding the 2008 financial crisis, however, given that the two indexes are built in a similar way, and that BFCI+ ranks noticeably worse than BFCI in the shorter sample, we exclude BFCI+ from the set of best-performing indexes. We replace BFCI+ with the Kansas City index, which ranks sixth in the 1995-2012 sample, and fifth when excluding the years around the 2008 financial crisis. In the second step we form all combinations of the five indexes selected above,11 calculate each combination’s first principal component, and regress12 changes in the first principal component of the FCIs (out of the 12 we study) that are not in the combination under consideration on changes in the first principal component of the combination. Letting C denote the combination of interest: PC = fpc({FCI } ) ∈/C j j∈/C PC = fpc({FCI } ) ∈C j j∈C ∆PC = γ +δ ×∆PC +ε ∈/C ∈C t In order to minimize the risk of overfitting, the regressions are run on several subsamples, and we use the resulting set of adjusted R2 to select the “best” combination of FCIs. Specifically, we calculate, for each combination and in each subsample, the squared deviation of the combination’s adjusted R2 relative to the highest adjusted R2 in each sub- 11 We form 31 different combinations: five individual indexes, ten sets of two indexes, ten sets of three indexes, five sets of four indexes, and one set of five indexes. 12 We again use robust regressions (see Hamilton (1991)). 13

sample. We then average, for each combination, the squared deviations across time periods, and use the averages to identify the “best” composite FCI. Table 15 reports five averages: the first (column A) shows arithmetic averages; in the second (B) the average is weighted by the ratio of daily S&P 500 return volatility in each subsample over the volatility in the full sample; in the third (C) it is weighted by the ratio of the average VIX level in each subsample over the average VIX level in the full sample; in the fourth (D) weights are based on the volatility of daily VIX changes; in the fifth (E) the arithmetic average is calculated on the four non-overlapping samples (7/95-12/98 through 1/06-6/12). The criterion we use to rank the FCIs is, of course, one of potentially many. For example, we could have selected the FCIs with the lowest volatility. Such choice would have implied an assumption on the way financial conditions change over time, namely that they evolve smoothly. Choosing the FCIs with the highest volatility would have implied that we assume financial conditions can change rapidly, and that we are looking for a more reactive proxy. Precisely to avoid imposing strong assumptions on the nature of the process for financial conditions, we have adopted a criterion that focuses on making efficient use of the available data, and that only assumes that (1) all the indexes we study provide some information about financial conditions, and that (2) none of the indexes is likely to contain uniquely accurate information about financial conditions. The results in Table 15 show that the first principal component of the St. Louis Fed, Bloomberg, Chicago Fed, and Citi indexes has the lowest average squared deviations in all columns (A) to (E): hence we consider such combination as our Composite FCI (CFCI). Figure 4 highlights that the CFCI follows the general pattern of the individual FCIs, but its volatility exhibits different regimes depending on whether financial conditions are loose or tight. The three panels of Figure 5 show the CFCI against three alternative proxies for financial conditions: the St. Louis Fed index, which is the best performing individual FCI, the first principal component of the St. Louis Fed and of the index with the lowest correlation with STLFSI (the MS FCI), and the implied volatility index VIX, which is one 14

of the variables underlying many FCIs. The CFCI tracks the STLFSI closely, although the latter is less volatile in the years following the bull market of the late 1990s. The first principal component of STLFSI and of MS FCI is more volatile than the CFCI, especially in the earlier part of the sample, and it points to much more improved conditions than the CFCI in early 2008, just before the crisis gained full traction. A comparison of VIX and the CFCI shows that the two track each other quite well, with the exception of the period between mid-2007 and late 2008, when VIX remains stable, and the CFCI shows a largely steady deterioration in financial conditions. In addition, the CFCI points to loose financial conditions in the second half of the 1990s until late 1998, while, over the same period, VIX points to slowly deteriorating conditions starting in 1995. In Figure 6 we plot the 24-month exponentially-weighted rolling correlation between VIX changes and changes in the CFCI, where observations are weighted so that the weight decays by 50% every 12 months (see Figure 6 for details). The correlation is initially low, but it jumps to about 80% with the Russian default in August 1998. With the exception of two relatively short periods in late 2000 and 2005/2006, it stays mostly above 50%, and it has been around 80-90% since the events of the late summer of 2008. 4 Conclusion We provide an assessment of the one-month and one-quarter ahead predictive power that a selection of indexes of financial conditions and financial stress (to which we collectively refer as FCIs) have for returns on a broad equity index and a set of equity industry portfolios, and for innovations to log-changes in macroeconomic variables that proxy for the state of consumer and business credit, manufacturing, and housing. The evidence for predictive power at the horizons we consider is weak – unless the financial crisis is included – and it highlights the role of non-synchronous trading and, potentially, data mining. 15

We also suggest a procedure for combining the various FCIs into a single proxy for financial conditions, which summarizes the different variable sets and aggregation methods used in building the individual FCIs. The composite FCI follows the pattern of the individual FCIs, but it clearly exhibits different volatility regimes according to whether financial conditions are tighter or looser than the historical norm, a feature that can be useful when evaluating the current state of financial conditions. 16

References Atchison, M., K. Butler, and R. Simonds, 1987, “Nonsynchronous Security Trading and Market Index Autocorrelation,” Journal of Finance, 42(1), 111–118. Bernanke, B., 1983, “Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression,” American Economic Review, 73(3), 257–276. Bernanke, B., and A. Blinder, 1992, “The Federal Funds Rate and the Channels of Monetary Transmission,” American Economic Review, 82(4), 901–921. Bollerslev, T., G. Tauchen, and H. Zhou, 2009, “Expected stock returns and variance risk premia,” Review of Financial Studies, 22(11), 4463–4492. Brave, S., and R. Butters, 2010, “Gathering Insights on the Forest from the Trees: A New Metric for Financial Conditions,” Federal Reserve Bank of Chicago, Working Paper, (WP- 2010-07). Calomiris, C., and J. Mason, 2003, “Consequences of Bank Distress During the Great Depression,” American Economic Review, 93(3), 937–947. Campbell, J., and M. Yogo, 2006, “Efficient tests of stock return predictability,” Journal of Financial Economics, 81(1), 27–60. Campello, M., J. Graham, and C. Harvey, 2010, “The real effects of financial constraints: Evidences from a financial crisis,” Journal of Financial Economics, 97(3), 470–487. Carlson, M., K. Lewis, and W. Nelson, 2012, “Using Policy Intervention to Identify Financial Stress,” Federal Reserve Board, Finance and Economics Discussion Series, 2. Chava, S., and A. Purnanandam, 2011, “The effect of banking crisis on bank-dependent borrowers,” Journal of Financial Economics, 99(1), 116–135. D’Antonio, P., 2008, “Appenix,” EMA Special Report, Citigroup Global Markets Inc., pp. 26–28. Gertler, M., and S. Gilchrist, 1994, “Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms,” Quarterly Journal of Economics, 109(2), 309–340. Hakkio, C., and W. 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, (Q2), 5–50. Hamilton, L., 1991, “srd1: How robust is robust regression?,” Stata Technical Bulletin, 2, 21–26. Hatzius, J., P. Hooper, F. Mishkin, K. Schoenholtz, and M. Watson, 2010, “Financial Conditions Indexes: A Fresh Look After the Financial Crisis,” NBER Working Paper 16150. 17

Hubrich, K., and R. Tetlow, 2012, “Financial Stress and Economic Dynamics: the transmission of crises,” Working Paper. Kashyap, A., O. Lamont, and J. Stein, 1994, “Credit Conditions and the Cyclical Behavior of Inventories,” Quarterly Journal of Economics, 109(3), 565–592. Kashyap, A., and J. Stein, 2000, “What Do a Million Observations on Banks Say About the Transmission of Monetary Policy,” American Economic Review, 90(3), 407–428. Kashyap, A., J. Stein, and D. Wilcox, 1993, “Monetary Policy and Credit Conditions: EvidencefromtheCompositionofExternalFinance,”American Economic Review,83(1),78–98. Khwaja, A., and A. Mian, 2008, “Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market,” American Economic Review, 98(4), 1413–1442. Kliesen, K., M. Owyang, and E. Vermann, 2012, “Disentangling Diverse Measures: a Survey of Financial Stress Indexes,” Federal Reserve Bank of St. Louis Review, 94(5), 369–397. Kliesen, K., andD.Smith, 2010, “MeasuringFinancialMarketStress,” Federal Reserve Bank of St. Louis, Economic Synopses, (2). Matheson, T., 2011, “Financial Conditions Indexes for the United States and Euro Area,” International Monetary Fund, Working Paper. Oet, M., R. Eiben, T. Bianco, D. Gramlich, and S. Ong, 2011, “The Financial Stress Index: Identification of Systemic Risk Conditions,” Federal Reserve Bank of Cleveland, Working Paper, 11-30. Paravisini, D., 2008, “Local Bank Financial Constraints and Firm Access to External Finance,” Journal of Finance, 63(5), 2161–2193. Peek, J., and E. Rosengren, 1997, “The International Transmission of Financial Shocks: The Case of Japan,” American Economic Review, 87(4), 495–505. , 2000, “Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States,” American Economic Review, 90(1), 30–45. Rosenberg, M. R., 2009, “Global Financial Market Trends and Policy,” Bloomberg Financial Conditions Watch, 2(6). 18

Figure 1: Time series of the TED spread. The graph shows the time series of the spread, in basis points, between the three-month USD LIBOR and the three month U.S. Treasury yield, which is commonly referred to as “TED spread”. Data are from the British Bankers’ Association and the Federal Reserve H.15 Statistical Release. The vertical lines highlight five events that range from the Asian crisis in 1997 to the European debt crisis in 2010. 19

Figure 2: Time series plots of the 12 financial conditions indexes. The three panels show the time series of the 12 FCIs we study. For scale reasons, the IMF U.S. Financial Stress Index and the natural logarithm of the Carlson, Lewis, and Nelson (2012) Financial Stress Index are shown separately in, respectively, the middle and bottom panels. See Table 1 for a list of FCI acronyms. 20

Figure 3: Threshold effects. The plot shows a scatter of innovations to log-changes in total inventories against the St. Louis Fed’s Financial Stress Index. The green squares and red triangles correspond to observations for which the index is above its 80th percentile, with the red triangles indicating an observation from between 2007 and 2012 and the green squares indicating an observation from oustide that range. July 1995 to June 2012. Figure 4: The Composite FCI. The Composite FCI is the first principal component of four indexes: STLFSI, BFCI, NFCI, and Citi FCI. The four indexes are selected on the basis of the procedure described in Section 3. 21

Figure 5: The Composite FCI and other proxies of financial conditions. EachofthethreepanelsshowtheCompositeFCI(CFCI,blueline)againstoneofthreealternativeproxiesfor financial conditions: the St. Louis Fed index (top panel), which is the individual FCI that best summarizes the information in the remaining FCIs (see Table 14); the first principal component of the St. Louis Fed index and of the index that is least correlated with the St. Louis Fed index (the Morgan Stanley FCI – middle panel); and the implied volatility index VIX (bottom panel). VIX data are from the CBOE. 22

Figure 6: Exponentially-weighted correlation between VIX changes and changes in the Composite FCI. The graph shows the exponentially-weighted correlation (black dotted line) between monthly changes in the volatility index VIX and monthly changes in the Composite FCI, together with the VIX index, which is standardized for scale reasons (red dashed line), and the Composite FCI (CFCI, blue solid line). The correlation is calculated on the basis of a 24-month rolling window, where the weights decay by 50% every 12months. Specifically, theweightsassignedtoobservations{t−i}23 aregivenby: 1 ·α·(1−α)i, where i=0 0.75 −ln(4) α=1−e 24 . 23

.sICF 21 eht fo snoitpircsed yrammuS :1 elbaT rehtehw etacidni nmuloc ”noitagerggA & ycneuqerF“ eht ni snoitaiverbba ehT .yduts ew sICF 21 eht no noitamrofni yrammus sedivorp elbat ehT dethgiew a htiw detagergga era xedni eht gniylrednu selbairav eht rehtehw dna ,)M( ylhtnom ro ,)W( ylkeew ,)D( yliad si xedni eht fo ycneuqerf eht xedninanodesabsi)2102(nosleNdna ,siweL ,noslraCniICFehT.)MFD(ledomrotcaf cimanydro,)CPF(tnenopmoclapicnirptsrfi,)AW( egareva ew sisylana laciripme eht nI .ytilibaborp a si hcihw ,ISF NLC eht fo noitpecxe eht htiw ,serocs-z sa desserpxe era sexedni llA .3002 ni depoleved .)elpmas ruo ni orez yltcaxe si reven xedni eht( mhtiragol larutan sti htiw ISF NLC eht ecalper ecruoS ataD ecnerefeR raeY tratS & ycneuqerF trohS emaN lluF .tsE raeY noitagerggA emaN grebmoolB )9002( grebnesoR 9002 4991 AW-D ICFB xednI snoitidnoC laicnaniF .S.U grebmoolB grebmoolB )9002( grebnesoR 9002 4991 AW-D +ICFB sulP xednI snoitidnoC laicnaniF .S.U grebmoolB deF dnalevelC )1102( .la te teO 9002 1991 AW-D ISFC xednI ssertS laicnaniF dnalevelC grebmoolB A/N 5991 AW-D ICF SM .S.U xednI snoitidnoC laicnaniF yelnatS nagroM srohtuA )2102( .la te noslraC 3002 4991 CPF-W ISF NLC xednI ssertS tekraM laicnaniF deF ogacihC )0102( srettuB dna evarB 6002 3791 CPF-W ICFN xednI snoitidnoC laicnaniF lanoitaN deF ogacihC )0102( srettuB dna evarB 6002 3791 CPF-W ICFNA xednI snoitidnoC laicnaniF lanoitaN detsujdA deF siuoL .tS )0102( htimS dna neseilK 9002 3991 CPF-W ISFLTS xednI ssertS laicnaniF deF siuoL .tS deF ytiC sasnaK )9002( noteeK dna oikkaH 9002 0991 CPF-M ISFCK xednI ssertS laicnaniF ytiC sasnaK hcraeseR itiC )8002( oinotnA’D 0002 3891 AW-M ICF itiC xednI snoitidnoC laicnaniF itiC rohtuA )1102( nosehtaM 9002 4991 MFD-M ICF FMI xednI snoitidnoC laicnaniF .S.U .F.M.I srohtuA 1102 .la te illeradraC 1102 0891 AW-M ISF FMI xednI ssertS laicnaniF .S.U .F.M.I 24

Table 2: Summary statistics of the 12 FCIs. The table shows the mean, median, 10th and 90th percentiles, peak crisis value (September 2008), total observations,and90%confidenceintervalfortheAR(1)autoregressiveroot(seeCampbellandYogo(2006)). July 1995 to June 2012. Daily Value on 90% CI for Mean Median 10th 90th Sept. 2008 # obs AR root BFCI 0.426 0.060 -0.887 1.910 7.651 4435 0.922 0.996 BFCI+ 0.264 -0.057 -1.237 1.782 5.980 4435 0.901 0.986 CFSI 0.138 0.024 -1.065 1.613 2.592 4435 0.939 1.004 MS FCI 0.075 -0.139 -1.144 1.718 0.764 4429 0.834 0.947 Weekly Value on 90% CI for Mean Median 10th 90th Sept. 2008 # obs AR root CLN FSI -4.311 -4.625 -7.264 -0.660 -0.001 887 0.923 0.997 NFCI -0.361 -0.534 -0.783 0.229 1.847 887 0.936 1.002 ANFCI -0.030 -0.224 -0.815 0.923 3.310 887 0.757 0.898 STLFSI 0.050 -0.127 -0.905 0.909 2.904 887 0.936 1.002 Monthly Value on 90% CI for Mean Median 10th 90th Sept. 2008 # obs AR root KCFSI 0.132 -0.115 -0.800 1.010 2.730 204 0.945 1.006 Citi FCI 0.137 0.013 -1.172 1.694 3.224 204 0.916 0.993 IMF FCI 0.078 -0.189 -0.781 1.100 2.930 204 0.974 1.015 IMF FSI -0.216 -1.065 -3.425 3.393 8.930 204 0.807 0.930 25

.snoitalerroc ICF ylhtnoM :3 elbaT ni noitavresbo tsetal eht ekat ew sICF yliad dna ylkeew roF .ycneuqerf ylhtnom ta derusaem ,sICF 21 eht gnoma snoitalerroc raenil swohs elbat ehT .2102 enuJ ot 5991 yluJ .htnom radnelac a ISFLTS ISF NLC ICFN ICF SM ICF FMI ISFCK ISF FMI ISFC ICF itiC +ICFB ICFB ICFNA 001 ICFNA 001 44.06 ICFB 001 13.09 58.46 +ICFB 001 76.56 84.87 98.32 ICF itiC 001 24.96 34.86 28.97 14.15 ISFC 001 02.07 79.66 87.87 38.28 97.56 ISF FMI 001 89.88 35.87 90.67 06.09 59.29 42.06 ISFCK 001 75.39 47.87 42.57 54.67 15.19 53.98 61.15 ICF FMI 001 20.54 30.83 81.83 58.23 50.93 38.54 36.64 90.22 ICF SM 001 23.14 90.29 17.59 95.88 20.97 42.27 26.98 67.39 19.76 ICFN 001 25.27 56.84 59.77 05.47 79.26 69.08 99.86 15.96 82.67 14.83 ISF NLC 001 70.67 57.19 07.93 84.09 49.29 92.97 15.47 54.37 88.98 75.19 32.94 ISFLTS 26

Table 4: Summary statistics of industry stock returns and log-changes of macro variables. The table reports summary statistics for the returns on the S&P 500 index and on industry portfolios, and forlog-changesinselectedmacroeconomicvariables. Theindustryportfoliosareequally-weighted, andbuilt using CRSP data on the basis of SIC code (finance: 6000 to 6231 and 6712 to 6726; construction: 1521 to 1799; manufacturing: 2011 to 3999; transportation: 4011 to 4971; wholesale trade: 5012 to 5199; retail trade: 5211 to 5999; services: 7011 to 8999). The macroeconomic variables are (with the St. Louis Fed’s FRED mnemonic in parentheses): commercial and industrial loans at all commercial banks (BUSLOANS), total consumer credit owned and securitized (TOTALSL), manufacturers’ new orders for durable goods (DGORDER),housingstarts(HOUST),industrialproduction(INDPRO),andvalueofmanufacturers’total inventories (AMTMTI). July 1995 to June 2012. Stock returns Mean Std. Dev. Median 10th 90th # obs S&P 500 0.006 0.046 0.010 -0.060 0.059 204 Finance 0.009 0.046 0.012 -0.033 0.059 204 Constr. 0.012 0.080 0.016 -0.085 0.100 204 Manuf. 0.012 0.073 0.015 -0.081 0.099 204 Transp. 0.009 0.061 0.017 -0.071 0.072 204 Whol. Trade 0.012 0.067 0.020 -0.070 0.085 204 Ret. Trade 0.011 0.072 0.009 -0.066 0.078 204 Services 0.012 0.084 0.020 -0.086 0.101 204 Macro variables Mean Std. Dev. Median 10th 90th # obs C&I 0.004 0.010 0.005 -0.011 0.015 204 Cons. Credit 0.005 0.006 0.004 -0.003 0.009 204 Dur. Goods 0.002 0.042 0.003 -0.047 0.049 204 Hous. Starts -0.003 0.069 -0.007 -0.089 0.080 204 Ind. Prod. 0.002 0.007 0.002 -0.006 0.009 204 Tot. Invent. 0.002 0.007 0.002 -0.008 0.010 204 27

Table 5: Predictive regressions: monthly stock returns, 1995-2012. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-month ahead stock returns on FCI levels. See Tables 1 and 4 for more details on the FCIs and for the definition of industry portfolios. Asterisks indicate significance at the 90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to June 2012. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -0.432* 4.604 -0.288 4.629 -0.180 4.649 0.022 4.652 Finance -0.624* 4.517 -0.420* 4.569 -0.778* 4.550 -0.241* 4.611 Constr. -0.164 8.079 -0.054 8.082 0.291 8.077 -0.227 8.079 Manuf. -0.263 7.265 -0.055 7.276 0.186 7.274 -0.415* 7.262 Transp. -0.315 6.122 -0.119 6.138 0.005 6.141 -0.369* 6.127 Whol. Trade -0.303 6.676 -0.100 6.690 0.115 6.691 -0.301* 6.684 Ret. Trade -0.194 7.227 0.029 7.233 0.114 7.232 0.016 7.233 Services -0.108 8.393 0.111 8.393 0.481 8.380 -0.182 8.392 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -1.239* 4.596 -0.069 4.652 -0.511 4.622 -0.117 4.644 Finance -1.843* 4.491 -1.068* 4.545 -0.660* 4.567 -0.271* 4.573 Constr. -0.500 8.077 -0.205 8.081 0.230 8.079 -0.027 8.082 Manuf. -0.575 7.269 0.020 7.276 0.076 7.276 0.035 7.276 Transp. -0.840 6.121 0.042 6.141 -0.116 6.139 -0.057 6.139 Whol. Trade -0.697 6.680 -0.368 6.686 0.080 6.692 0.074 6.690 Ret. Trade -0.223 7.232 -0.417 7.226 0.444 7.218 0.100 7.229 Services -0.393 8.391 0.279 8.392 0.275 8.390 0.179 8.384 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -0.583* 4.608 -0.566 4.608 -0.920* 4.565 -0.041 4.650 Finance -0.766* 4.542 -0.406 4.595 -0.770* 4.557 -0.200* 4.558 Constr. 0.063 8.082 0.030 8.083 -0.170 8.081 0.078 8.077 Manuf. -0.013 7.276 -0.007 7.276 -0.328 7.269 0.039 7.275 Transp. -0.201 6.137 -0.281 6.132 -0.587 6.114 0.000 6.141 Whol. Trade -0.048 6.692 -0.060 6.692 -0.186 6.690 0.000 6.692 Ret. Trade 0.282 7.226 -0.071 7.232 0.167 7.231 0.107 7.222 Services 0.130 8.393 -0.122 8.393 -0.209 8.392 0.078 8.389 28

Table 6: Predictive regressions: monthly stock returns excluding the first day of the month, 1995-2012. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-month ahead stock returns on FCI levels. Returns are calculated after excluding the first day of each month. SeeTables1and4formoredetailsontheFCIsandforthedefinitionofindustryportfolios. Asterisks indicate significance at the 90% level. Statistical significance is evaluated on the basis of heteroskedasticityconsistentstandarderrors,or,whereappropriate,withthelocal-to-unityasymptoticsprocedureofCampbell and Yogo (2006) (see Section 2 for details). July 1995 to June 2012. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -0.230 4.620 -0.110 4.630 -0.027 4.633 0.308 4.621 Finance -0.453* 4.581 -0.263 4.615 -0.634* 4.589 0.007 4.634 Constr. 0.071 8.295 0.198 8.289 0.438 8.283 0.192 8.293 Manuf. -0.060 7.146 0.124 7.144 0.314 7.139 -0.039 7.146 Transp. -0.156 6.067 0.013 6.072 0.117 6.070 -0.057 6.071 Whol. Trade -0.091 6.529 0.091 6.529 0.238 6.526 0.086 6.530 Ret. Trade 0.062 7.190 0.247 7.180 0.278 7.185 0.448 7.174 Services 0.075 8.149 0.273 8.139 0.579 8.129 0.216 8.147 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -0.670 4.617 0.113 4.633 -0.193 4.629 -0.041 4.632 Finance -1.349* 4.567 -0.922* 4.580 -0.386 4.616 -0.197 4.610 Constr. 0.188 8.295 0.039 8.295 0.602 8.271 0.078 8.293 Manuf. -0.081 7.146 0.131 7.146 0.355 7.137 0.121 7.140 Transp. -0.422 6.067 0.129 6.071 0.103 6.071 0.008 6.072 Whol. Trade -0.186 6.530 -0.199 6.529 0.361 6.520 0.147 6.521 Ret. Trade 0.364 7.187 -0.175 7.189 0.762 7.147 0.187 7.177 Services 0.042 8.150 0.369 8.145 0.521 8.132 0.258 8.127 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -0.329 4.619 -0.318 4.619 -0.633* 4.592 0.050 4.630 Finance -0.540* 4.596 -0.260 4.625 -0.547 4.603 -0.121 4.612 Constr. 0.360 8.286 0.189 8.293 0.160 8.294 0.183 8.267 Manuf. 0.192 7.143 0.173 7.144 -0.085 7.146 0.125 7.131 Transp. -0.041 6.071 -0.139 6.070 -0.399 6.059 0.067 6.066 Whol. Trade 0.159 6.528 0.089 6.530 0.054 6.531 0.078 6.524 Ret. Trade 0.531 7.167 0.155 7.188 0.424 7.179 0.204 7.151 Services 0.296 8.144 0.048 8.150 0.003 8.150 0.147 8.132 29

Table 7: Predictive regressions: monthly stock returns, 1995-2006. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-month ahead stock returns on FCI levels. See Tables 1 and 4 for more details on the FCIs and for the definition of industry portfolios. Asterisks indicate significance at the 90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to December 2006. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 0.147 4.277 0.038 4.279 0.718 4.245 0.436 4.253 Finance 0.180 3.392 0.279 3.384 -0.040 3.395 0.512 3.350 Constr. 0.524 6.737 0.057 6.751 1.080 6.703 -0.027 6.751 Manuf. 0.979 6.980 0.351 7.019 1.739 6.907 0.045 7.028 Transp. 0.625 5.983 0.209 6.002 1.291 5.928 0.119 6.004 Whol. Trade 0.715 6.170 0.210 6.196 1.192 6.135 0.310 6.190 Ret. Trade 0.444 6.140 0.154 6.149 0.911 6.113 0.707 6.104 Services 1.474 8.853 0.861 8.895 2.400 8.757 0.327 8.932 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -0.070 4.279 1.117* 4.228 -0.331 4.275 0.045 4.278 Finance 0.214 3.395 -0.547 3.380 0.889 3.366 0.156 3.382 Constr. 1.120 6.746 0.096 6.751 1.149 6.726 0.242 6.735 Manuf. 3.335 6.981 0.856 7.010 1.724 6.974 0.434 6.978 Transp. 1.169 5.999 1.027 5.975 0.773 5.993 0.242 5.988 Whol. Trade 1.695 6.185 0.053 6.199 1.529 6.151 0.428 6.144 Ret. Trade 1.015 6.146 -0.107 6.151 1.347 6.113 0.389 6.105 Services 3.748 8.891 1.542 8.892 2.440 8.853 0.732 8.825 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -0.493 4.268 -0.451 4.253 -0.929 4.239 0.177 4.260 Finance -0.029 3.395 0.218 3.388 0.590 3.375 -0.069 3.392 Constr. 0.421 6.746 0.089 6.751 0.265 6.749 0.029 6.751 Manuf. 1.005 7.001 0.401 7.016 0.396 7.024 0.267 7.003 Transp. 0.373 6.001 -0.063 6.005 -0.371 6.001 0.248 5.980 Whol. Trade 0.575 6.189 0.222 6.195 0.606 6.188 0.084 6.196 Ret. Trade 0.479 6.144 -0.020 6.151 0.838 6.129 0.029 6.151 Services 1.252 8.905 0.145 8.937 0.560 8.932 0.369 8.900 30

Table 8: Predictive regressions: quarterly stock returns, 1995-2012. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-quarter ahead stock returns on FCI levels. See Tables 1 and 4 for more details on the FCIs and for the definition of industry portfolios. Asterisks indicate significance at the 90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to June 2012. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -1.099* 8.882 -0.668 8.989 -0.650* 9.027 0.800 9.005 Finance -1.401* 9.046 -0.960 9.188 -2.191* 9.043 0.041 9.314 Constr. -0.006 15.616 0.203 15.612 1.144 15.572 0.481 15.606 Manuf. -0.025 15.042 0.565 15.016 1.181 14.994 0.436 15.034 Transp. -0.498 12.529 0.165 12.551 0.141 12.553 0.196 12.552 Whol. Trade 0.179 14.199 0.714 14.156 0.919 14.171 1.210 14.135 Ret. Trade 0.620 14.885 1.278 14.778 1.101 14.875 1.897 14.760 Services 0.462 17.074 1.090 17.002 1.927 16.977 2.035 16.932 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -3.647* 8.787 -0.946 9.019 -1.171 8.972 -0.175 9.042 Finance -4.615* 8.900 -3.325* 8.917 -1.387 9.206 -0.611 9.197 Constr. -1.267 15.597 -1.657 15.558 1.217 15.566 0.432 15.581 Manuf. -0.726 15.036 -1.195 15.011 1.479 14.966 0.567 14.980 Transp. -2.066 12.493 -1.208 12.515 0.559 12.540 0.187 12.545 Whol. Trade -0.311 14.200 -1.626 14.140 1.882 14.071 0.713 14.097 Ret. Trade 1.256 14.898 -1.265 14.882 3.085 14.580 0.871 14.769 Services -0.701 17.085 -1.424 17.051 1.947 16.974 0.981 16.926 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -1.467* 8.918 -1.408 8.903 -2.324* 8.756 -0.078 9.047 Finance -1.665* 9.147 -0.630 9.286 -1.743 9.154 -0.494 9.148 Constr. 0.602 15.603 1.202 15.553 -0.173 15.615 0.232 15.594 Manuf. 1.092 14.998 1.201 14.978 0.034 15.042 0.384 14.981 Transp. 0.054 12.553 0.050 12.553 -0.921 12.521 0.206 12.532 Whol. Trade 1.456 14.118 1.122 14.142 0.653 14.187 0.448 14.113 Ret. Trade 2.559 14.671 1.309 14.839 1.777 14.813 0.650 14.739 Services 1.339 17.032 0.861 17.061 0.235 17.088 0.440 17.019 31

Table 9: Predictive regressions: quarterly stock returns, 1995-2006. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-quarter ahead stock returns on FCI levels. See Tables 1 and 4 for more details on the FCIs and for the definition of industry portfolios. Asterisks indicate significance at the 90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to December 2006. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 0.375 8.351 0.168 8.355 1.122 8.316 2.058* 7.984 Finance 1.338 6.788 1.173 6.776 -0.921* 6.853 1.698* 6.579 Constr. 3.094 13.512 1.314 13.707 2.639 13.638 1.062 13.717 Manuf. 4.004 13.913 1.732 14.225 5.299 13.799 1.906 14.157 Transp. 2.077 11.938 1.020 12.025 2.769 11.899 1.734 11.893 Whol. Trade 3.342 12.132 1.416 12.385 2.961 12.281 2.681* 12.052 Ret. Trade 3.199 12.507 1.841 12.665 2.178 12.710 3.560* 12.080 Services 5.677 17.299 3.557 17.595 7.327 17.157 3.757* 17.412 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -3.303 8.308 -1.628 8.301 -0.956 8.342 0.257 8.341 Finance 1.664 6.871 -2.247 6.756 3.035 6.705 0.398 6.841 Constr. 2.746 13.755 -2.492 13.696 4.777 13.552 0.958 13.644 Manuf. 5.456 14.262 -1.420 14.315 6.194 13.977 1.710 13.933 Transp. -1.532 12.065 -0.722 12.065 2.813 11.985 0.990 11.912 Whol. Trade 2.633 12.452 -2.846 12.358 5.263 12.172 1.509 12.109 Ret. Trade 2.919 12.786 -2.431 12.730 4.964 12.551 1.540 12.442 Services 4.453 17.944 -1.607 17.960 8.253 17.471 2.613* 17.222 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI S&P 500 -1.730 8.284 -1.548 8.186 -2.812 8.161 0.073 8.355 Finance 0.438 6.881 0.939 6.810 2.134 6.750 -0.399 6.814 Constr. 1.901 13.723 1.388 13.693 1.514 13.742 -0.299 13.755 Manuf. 2.813 14.229 1.877 14.195 1.485 14.309 0.197 14.331 Transp. 0.500 12.069 0.076 12.073 -0.890 12.059 0.095 12.070 Whol. Trade 1.673 12.428 1.257 12.398 1.710 12.425 -0.125 12.469 Ret. Trade 1.745 12.763 0.851 12.778 2.553 12.707 -0.185 12.803 Services 3.291 17.864 1.385 17.922 1.847 17.947 0.428 17.953 32

Table 10: Predictive regressions: innovations to monthly log-changes in macroeconomic variables, 1995-2012. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-month ahead innovations to log-changes in macroeconomic variables on FCI levels. See Tables 1 and 4 formoredetailsontheFCIsandonthemacroeconomicvariables. Asterisksindicatesignificanceatthe90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to June 2012. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.078* 0.662 0.105* 0.652 0.072 0.669 -0.051 0.671 Cons. Credit -0.034 0.519 -0.032 0.519 -0.047 0.519 0.018 0.521 Dur. Goods -0.883* 3.733 -0.765* 3.781 -0.920* 3.860 -0.667* 3.905 Hous. Starts -1.184* 6.241 -0.895* 6.344 -0.759* 6.456 -0.764* 6.448 Ind. Prod. -0.120* 0.645 -0.094* 0.654 -0.167* 0.649 -0.126* 0.657 Tot. Invent. -0.059* 0.437 -0.054* 0.438 -0.043 0.444 -0.071* 0.439 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.240* 0.658 0.059 0.672 0.180* 0.647 -0.038 0.667 Cons. Credit -0.129* 0.516 -0.025 0.521 -0.073* 0.516 -0.017 0.520 Dur. Goods -2.366* 3.725 -1.147* 3.875 -1.153* 3.786 -0.317* 3.900 Hous. Starts -2.984* 6.265 -1.875* 6.342 -1.220* 6.376 -0.321* 6.457 Ind. Prod. -0.321* 0.645 -0.051 0.670 -0.162* 0.650 0.072* 0.649 Tot. Invent. -0.143* 0.438 -0.050 0.444 -0.073* 0.439 -0.021 0.443 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.168* 0.647 0.119* 0.660 0.183* 0.649 0.034 0.661 Cons. Credit -0.061* 0.517 -0.034 0.520 -0.068* 0.517 -0.013 0.519 Dur. Goods -1.247* 3.731 -0.834* 3.860 -1.356* 3.746 -0.335 3.772 Hous. Starts -1.282* 6.349 1.162* 6.369 -1.390* 6.359 -0.414 6.317 Ind. Prod. 0.177* 0.643 0.180* 0.640 -0.213* 0.638 -0.044 0.651 Tot. Invent. -0.071* 0.439 -0.075* 0.438 -0.106* 0.434 -0.017 0.442 33

Table 11: Predictive regressions: innovations to monthly log-changes in macroeconomic variables, 1995-2006. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-month ahead innovations to log-changes in macroeconomic variables on FCI levels. See Tables 1 and 4 formoredetailsontheFCIsandonthemacroeconomicvariables. Asterisksindicatesignificanceatthe90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to December 2006. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I -0.085 0.583 -0.087* 0.580 0.042 0.586 -0.041 0.586 Cons. Credit 0.030 0.332 0.040 0.331 0.005 0.333 -0.029 0.332 Dur. Goods 0.339 3.683 0.309 3.681 0.420 3.681 -0.170 3.690 Hous. Starts 0.225 5.249 0.382 5.238 0.414 5.244 0.455 5.230 Ind. Prod. 0.138* 0.547 0.059 0.556 -0.092 0.555 -0.082* 0.552 Tot. Invent. -0.040 0.383 -0.052 0.380 -0.022 0.384 -0.075* 0.375 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.217 0.585 -0.058 0.586 -0.221* 0.576 -0.057* 0.577 Cons. Credit -0.118 0.332 -0.058 0.332 -0.016 0.333 0.015 0.332 Dur. Goods 2.128 3.657 0.179 3.693 0.667 3.679 -0.185 3.677 Hous. Starts 0.465 5.251 0.120 5.252 -0.827 5.236 0.145 5.245 Ind. Prod. 0.363* 0.552 -0.054 0.559 0.153 0.554 0.058* 0.548 Tot. Invent. -0.179 0.382 0.048 0.383 -0.098 0.381 -0.033* 0.379 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.128 0.582 0.076 0.582 0.227* 0.570 0.009 0.587 Cons. Credit 0.058 0.332 0.021 0.333 0.055 0.332 0.020 0.330 Dur. Goods 0.876 3.655 -0.274 3.683 -0.807 3.659 0.187 3.670 Hous. Starts 0.144 5.252 -0.256 5.246 0.490 5.244 -0.091 5.249 Ind. Prod. 0.169* 0.550 0.132* 0.542 -0.147* 0.552 0.041 0.552 Tot. Invent. -0.079 0.381 0.052 0.380 -0.149* 0.373 -0.016 0.383 34

Table 12: Predictive regressions: innovations to quarterly log-changes in macroeconomic variables, 1995-2012. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-quarter ahead innovations to log-changes in macroeconomic variables on FCI levels. See Tables 1 and 4 formoredetailsontheFCIsandonthemacroeconomicvariables. Asterisksindicatesignificanceatthe90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to June 2012. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.406* 1.229 0.477* 1.161 0.571* 1.260 -0.414* 1.305 Cons. Credit -0.099 1.003 -0.122 0.996 -0.121 1.007 -0.029 1.014 Dur. Goods -1.858* 5.308 -1.637* 5.483 -1.413* 5.896 -1.450* 5.840 Hous. Starts -1.373* 8.645 -0.803 8.822 -1.077* 8.846 0.781 8.870 Ind. Prod. -0.248* 1.291 -0.167 1.324 -0.371* 1.296 -0.348* 1.290 Tot. Invent. -0.285* 1.112 -0.257* 1.128 -0.238* 1.175 -0.197 1.179 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 1.238* 1.176 0.548* 1.314 0.752* 1.155 0.204* 1.298 Cons. Credit -0.329 0.996 -0.064 1.014 -0.174 0.999 -0.014 1.014 Dur. Goods -4.935* 5.311 -2.643* 5.679 -2.278* 5.604 -0.708* 5.824 Hous. Starts -4.171* 8.562 -2.657 8.651 -0.784 8.878 -0.386 8.865 Ind. Prod. -0.657* 1.292 -0.061 1.349 -0.289* 1.317 -0.163* 1.291 Tot. Invent. -0.578* 1.150 0.420* 1.151 -0.258* 1.171 -0.101* 1.175 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.780* 1.117 0.526* 1.246 0.812* 1.132 0.201 1.191 Cons. Credit -0.145 1.003 -0.089 1.010 -0.208* 0.994 -0.038 1.006 Dur. Goods -2.415* 5.508 -1.635* 5.765 -2.866* 5.369 -0.635 5.637 Hous. Starts -1.163 8.829 -1.025 8.834 -2.011* 8.690 -0.375 8.815 Ind. Prod. -0.393* 1.285 -0.384* 1.274 -0.430* 1.281 -0.117 1.285 Tot. Invent. -0.282* 1.163 0.334* 1.136 -0.375* 1.141 -0.094 1.153 35

Table 13: Predictive regressions: innovations to quarterly log-changes in macroeconomic variables, 1995-2006. The table reports the coefficients and root mean squared errors (both in %) of predictive regressions of one-quarter ahead innovations to log-changes in macroeconomic variables on FCI levels. See Tables 1 and 4 formoredetailsontheFCIsandonthemacroeconomicvariables. Asterisksindicatesignificanceatthe90% level. Statistical significance is evaluated on the basis of heteroskedasticity-consistent standard errors, or, where appropriate, with the local-to-unity asymptotics procedure of Campbell and Yogo (2006) (see Section 2 for details). July 1995 to December 2006. BFCI BFCI+ CFSI MS FCI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.431* 1.111 0.399* 1.095 0.436 1.127 -0.292 1.118 Cons. Credit 0.058 0.607 0.028 0.609 0.094 0.606 -0.065 0.605 Dur. Goods 0.956 4.426 -0.960 4.389 0.748 4.469 -0.768* 4.408 Hous. Starts 1.513 5.349 1.758* 5.190 1.210 5.434 1.506* 5.203 Ind. Prod. -0.261 1.097 -0.096 1.115 -0.087 1.118 -0.216* 1.089 Tot. Invent. -0.364* 0.876 -0.374* 0.845 -0.232 0.915 -0.316* 0.851 NFCI ANFCI STLFSI CLN FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 1.375* 1.110 0.061 1.172 0.761* 1.104 -0.191* 1.109 Cons. Credit -0.352 0.602 -0.176 0.601 0.027 0.609 0.030 0.607 Dur. Goods 3.930 4.372 0.480 4.494 -1.337 4.450 -0.492* 4.396 Hous. Starts 2.044 5.478 0.501 5.499 2.546 5.347 -0.336 5.467 Ind. Prod. -0.693 1.103 -0.160 1.116 -0.310 1.108 -0.118 1.095 Tot. Invent. -1.189* 0.872 0.009 0.931 -0.625* 0.873 -0.165* 0.872 KCFSI Citi FCI IMF FCI IMF FSI β RMSE β RMSE β RMSE β RMSE FCI FCI FCI FCI C&I 0.699* 1.085 0.344* 1.111 0.810* 1.052 0.137 1.121 Cons. Credit 0.191 0.597 0.050 0.607 -0.066 0.608 -0.047 0.598 Dur. Goods 1.607 4.386 -0.413 4.481 -2.054* 4.308 -0.320 4.432 Hous. Starts 0.515 5.497 0.255 5.500 1.043 5.467 -0.085 5.503 Ind. Prod. -0.426* 1.087 0.362* 1.049 -0.303 1.103 -0.142 1.062 Tot. Invent. 0.523* 0.870 0.317* 0.865 -0.605* 0.847 0.142 0.861 36

Table 14: Explanatory power of FCIs. The table reports the adjusted R2s (in %) of robust regressions (Hamilton (1991)) of changes in the first principal component of all the FCIs (aside from the one indicated in each row) on changes in the indicated FCI (columns “∆PC”), and of one-lag autoregressive residuals of changes in the first principal component of all the FCIs (aside from the one indicated in each row) on one-lag autoregressive residuals of the changes in the indicated FCI (columns “Res.∆”). July 1995 - Dec. 2006 July 1995 - June 2012 ∆PC Res.∆ Average Rank ∆PC Res.∆ Average Rank BFCI 48.10 44.33 46.22 2 56.99 52.98 54.98 4 BFCI+ 34.99 30.76 32.88 6 58.35 54.20 56.27 3 Citi FCI 34.43 37.18 35.81 4 44.62 45.39 45.00 5 CFSI 32.49 29.44 30.97 24.70 21.89 23.30 NFCI 44.36 41.34 42.85 3 65.07 60.79 62.93 2 ANFCI 15.85 11.62 13.74 35.92 29.30 32.61 IMF FSI 22.33 27.99 25.16 33.35 35.60 34.48 KCFSI 35.44 31.61 33.52 5 42.23 36.41 39.32 6 IMF FCI 14.04 19.18 16.61 29.36 38.77 34.06 MS FCI 17.79 17.78 17.79 21.34 24.40 22.87 CLN FSI 29.96 27.70 28.83 33.61 30.96 32.28 STLFSI 63.72 60.01 61.86 1 75.63 73.82 74.72 1 37

snoitanibmoc ICF fo tnenopmoc lapicnirp tsrfi eht fo rewop yrotanalpxE :51 elbaT edisa( sICF eht lla fo tnenopmoc lapicnirp tsrfi eht ni segnahc fo ))1991( notlimaH( snoisserger tsubor fo 2R detsujda eht troper snmuloc xis tsrfi ehT era sICF laudividni rof stluser eht( noitanibmoc ICF detacidni eht fo tnenopmoc lapicnirp tsrfi eht ni segnahc no )wor hcae ni detacidni esoht morf hcae morf 2R detsujda s’noitanibmoc hcae fo snoitaived derauqs eht fo ,sdoirep-bus ssorca ,segareva eht wohs snmuloc evfi tsal ehT .)detalubatnu ,segareva dethgiew troper D hguorht B snmuloC .sdoirep-bus xis eht ssorca segareva elpmis swohs A nmuloC .2R detsujda tsegral s’doirep emit XIV xedni ytilitalov deilpmi egareva eht no ,)B( snruter 005 P&S yliad fo ytilitalov eht no desab era doirep-bus hcae ot dengissa sthgiew eht erehw ehthtiwdoirep-bushcaenisegareva/seitilitalovehtgnizilamronybdetaluclacerasthgiewehT.)D(XIVnisegnahcyliadfoytilitalovehtnodna,)C( .)21/6-60/1hguorht89/21-59/7(sdoirep-busgnippalrevo-nonruofehtssorcasegarevaelpmisswohsEnmuloC .elpmasllufehtrevoegareva/ytilitalov noitaiveD .qS egarevA -60/1 -20/1 -99/1 -59/7 -59/7 -59/7 E D C B A 21/6 50/21 10/21 89/21 60/21 21/6 noitanibmoC 08.1 34.1 64.1 14.1 84.1 96.67 18.94 44.36 47.65 86.85 63.07 ICFB - ISFLTS 41.1 97.0 48.0 08.0 48.0 86.18 27.35 42.36 63.06 71.26 21.67 ICFN - ISFLTS 11.3 93.2 94.2 14.2 14.2 20.17 08.36 77.25 55.15 14.75 72.07 ISFCK - ISFLTS 79.1 75.1 77.1 96.1 27.1 61.38 03.05 77.35 04.36 32.45 13.37 ICF itiC - ISFLTS 99.3 64.3 34.3 93.3 54.3 02.76 55.54 98.85 56.84 59.35 34.26 ICFN - ICFB 44.1 63.1 33.1 53.1 03.1 19.17 56.65 12.46 18.95 34.16 01.66 ISFCK - ICFB 62.2 96.1 19.1 58.1 08.1 43.77 55.15 52.15 94.56 39.75 69.07 ICF itiC - ICFB 43.5 77.4 58.4 08.4 67.4 64.36 41.74 47.84 19.84 23.84 30.16 ISFCK - ICFN 84.5 13.4 27.4 35.4 46.4 85.77 80.04 46.24 34.25 79.84 14.46 ICF itiC - ICFN 92.8 09.6 22.7 00.7 11.7 75.56 63.24 14.93 43.04 01.54 87.85 ISFCK - ICF itiC 21.1 09.0 88.0 88.0 19.0 40.67 99.05 38.07 88.06 69.16 23.17 ICFN - ICFB - ISFLTS 65.0 94.0 34.0 84.0 14.0 20.37 15.16 68.17 00.86 48.66 37.37 ISFCK - ICFB - ISFLTS 56.0 44.0 05.0 94.0 84.0 66.18 79.55 49.26 55.07 92.46 90.77 ICF itiC - ICFB - ISFLTS 53.2 58.1 19.1 68.1 58.1 81.27 33.36 51.65 66.45 23.95 66.96 ISFCK - ICFN - ISFLTS 40.1 96.0 38.0 97.0 77.0 70.68 61.55 26.75 73.27 22.26 13.87 ICF itiC - ICFN - ISFLTS 47.2 89.1 51.2 30.2 90.2 56.87 06.45 58.25 09.25 55.85 63.17 ISFCK - ICF itiC - ISFLTS 61.1 11.1 10.1 80.1 89.0 79.86 79.65 41.07 58.36 39.46 69.76 ISFCK - ICFN - ICFB 81.1 39.0 99.0 99.0 49.0 23.57 43.55 44.06 80.76 68.26 50.17 ICF itiC - ICFN - ICFB 99.0 86.0 87.0 67.0 27.0 56.97 40.75 75.85 61.27 29.36 98.47 ISFCK - ICF itiC - ICFB 35.3 76.2 29.2 08.2 38.2 78.57 59.74 97.84 86.55 15.45 25.96 ISFCK - ICF itiC - ICFN 16.0 55.0 84.0 35.0 64.0 97.17 81.95 01.67 11.96 44.86 08.27 ISFCK - ICFN - ICFB - ISFLTS 23.0 22.0 42.0 42.0 32.0 34.18 57.75 30.86 34.17 11.76 07.57 ICF itiC - ICFN - ICFB - ISFLTS 55.0 93.0 34.0 34.0 93.0 54.97 13.06 42.36 43.37 35.76 57.47 ISFCK - ICF itiC - ICFB - ISFLTS 84.1 20.1 51.1 90.1 90.1 48.08 71.75 37.65 22.16 62.26 82.47 ISFCK - ICF itiC - ICFN - ISFLTS 56.0 34.0 84.0 84.0 44.0 10.97 89.75 38.26 82.37 03.76 62.67 ISFCK - ICF itiC - ICFN - ICFB 63.0 82.0 82.0 92.0 62.0 55.77 21.06 64.86 54.27 69.86 54.57 ISFCK - ICF itiC - ICFN - ICFB - ISFLTS 38