Journal Issue

Euro Area Policies: Selected Issues

International Monetary Fund. European Dept.
Published Date:
July 2014
  • ShareShare
Show Summary Details

Investment in the Euro Area: Why Has It Been Weak?1

Investment across the euro area—both in real terms and in percent of GDP—remains below its precrisis level. Its performance has been weaker than in most previous recessions and financial crises. Staff analysis shows that much of this weakness can be explained by output dynamics, although a high cost of capital, financial constraints, corporate leverage, and uncertainty have also constrained investment in parts of the euro area. Investment is expected to pick up as the recovery strengthens and uncertainty declines. However, financial fragmentation and high corporate leverage in some countries will likely continue to weigh on investment.

A. Investment in the Euro Area: Post-crisis Trends

1. Investment has been hit hard since the onset of the crisis. It has not recovered since, including in many of the core economies. Total (real) investment remains below its pre-crisis level across the euro area.2 Part of this decline reflects declines in public and housing investment in certain countries. For example, housing investment declined from about 12–13 percent of GDP before the crisis to 6 percent of GDP in Spain and to 2-3 percent of GDP in Greece and Ireland after the crisis. Similarly, following the sovereign debt crisis, public investment, as well as private non-residential investment remains well below its pre-crisis levels, in most of the euro area, particularly in stressed countries.3

Investment Recovery to Date: 2013Q4

(Percent; 2007 quarterly average=100)

Sources: Eurostat; and staff calculations.

Private Non-Residential Investment Recovery to Date: 2013Q4

(Percent; 2007 quarterly average=100)

Sources: Eurostat, and staff calculations

Notes: 1/ DEU - last available data point 2012Q4; PRT = 2013Q3.

2. Overall the behavior of investment and its determinants vary largely across the euro area. Across countries, stressed countries have suffered more than the core countries. Across firm sizes, SMEs have suffered more than the larger corporations. And across sectors, services have suffered substantially.

Investment Growth Contributions

(Percentage points)

Sources: Eurostat; and IMF staff calculations.

3. Weak investment performance is associated with weak aggregate demand. Real GDP in the euro area remains below its pre-crisis level and is more sluggish than in typical recessions. While the recovery is gaining momentum, domestic demand growth is still fragile, and the output gap for the euro area is still negative and large. Given subdued aggregate demand, it is not surprising that investment has also lagged behind the trends observed in most previous recessions. Indeed, investment growth is still lower than real GDP growth in the euro area, unlike the recovery in the U.S.

Source: OECD; Eurostat; Staff calculations.

Notes: Quarterly data and yoy change. t=0 at 2008 Q1; the shade area starts from 2011 Q3 (t=15) to indicate the back-to-back recession in the euro area; historical episodes are based on CEPR-dated recessions: 1974Q3 to 1975Q1, 1980Q1 to 1982Q3, and 1992Q1 to 1993Q3.

Source: Eurostat; Staff calculations.

4. In addition, financial crises tend to have durable effects on investment, reflecting credit supply constraints, balance sheet problems, and other supply-side factors. Previous experience with financial crises shows that the decline in the investment-to-GDP ratio could be long-lasting, with a peak impact of 3 to 3½ percentage points three years after the crisis (WEO 2014). In the euro area, the observed decline in investment-to-GDP ratio since the beginning of the crisis is more severe than the standard financial crisis but is in line with the decline observed in the most severe crises—with the ratio standing at 4¼ percentage points below the pre-crisis level.

Source: WEO 2014, Chapter 3; and staff calculations.

Notes: Gross fixed capital formation in percent of GDP. The entire sample of advanced economy financial crises between 1970 and 2007 identified by Laeven and Valencia (2012). Big 5 financial crises are those in Spain, 1977; Norway, 1987; Finland, 1991; Sweden, 1991; and Japan, 1992. Dashed red lines denote 90 percent confidence bands; and black line denotes the actual evolution of the investment-to-GDP ratio in the euro area from 2007 to 2013. X-axis units are years; t = 0 denotes the year of the financial crisis.

5. In the euro area, the high cost of capital and limited access to funding could impose additional impediments to investment in certain countries. While the ECB’s policy rate is close to zero, lending rates remain elevated in some countries as financial fragmentation persists. Given that debt financing in the euro area is mostly bank-based (about 90 percent), this increases the cost of capital, particularly for smaller firms. In addition, many smaller companies have difficulty accessing credit (SAFE survey). Recent improvements in corporate bond and stock markets are likely to benefit only larger corporations.

Euro Area Corporate Lending Rates 1/


Sources: Haver, ECB, staff calculations.

1/ Unweighted average; MFI lending to corporations under €1 million, 1–5 years maturity. Core: Germany, France, Belgium, Netherlands. Stressed countries include Greece, Ireland, Italy, Portugal, and Spain. In the sample, Ireland is excluded from May 2011 and Greece from September 2012.

Outcome of Loan Application by EA Firms, 2013H2

(Weighted percentages of responses)

Source: ECB, Safe survey.

6. Moreover, the corporate sector is highly leveraged in some countries, further depressing credit flows to the private sector.4 Corporate debt-to-equity ratios remain elevated in some stressed countries, and the deleveraging process is still at an early stage. As companies repair their balance sheets and reduce debt, this feeds back into the banking sector through low demand for credit and higher NPLs. As a result of both corporate and banking sector deleveraging, the credit to the private sector continues to shrink.

Euro Area: NFC Debt/Equity Ratio 1/


Source: ECB.

1/ Debt at EA level is non-consolidated

Euro Area: Growth of Nominal Credit to Corporates

(Year-on-year percent change)

Source: Haver. The stock of credit for Slovenia includes the December 2013 transfer of nonperforming loans to BAMC.

7. Investment in the euro area could recover without credit, but credit-less recoveries are associated with lower investment and GDP growth. Empirically, credit-less recoveries are rare, especially in advanced economies, which suggests risks to recovery unless credit growth resumes (IMF, 2014). In addition such recoveries are associated with lower investment and output growth than in recoveries with credit. This, in turn, could have long-term consequences through lower potential output.

8. Against this backdrop, the paper explores to what extent output dynamics and other factors can explain private non-residential investment across the euro area. First, to analyze the impact of the output dynamics, an accelerator model is estimated for the euro area and selected euro area countries.5 While this model tracks investment closely, actual post-crisis investment has remained below its model-implied value for most countries. Second, to explore the impact of the cost of capital and financial constraints, we augment the model with the (real) cost of capital and a proxy for financial constraints (EC survey). These additional factors are significant for some of the countries; however, actual investment continues to remain below its estimated level for most countries. Finally, to explore the effects stemming from uncertainty, corporate leverage, and cash flow, a more eclectic (bond market) model is estimated. Controlling for these factors, changes in output are more representative of demand factors. Accordingly, uncertainty is associated with low investment in most countries. In addition, high corporate leverage is associated with subdued investment in Italy and Portugal, and to a lesser extent in Spain, and France. Overall, this model seems to be a better fit for stressed countries, with the residuals substantially smaller than in the previous two models.

B. Drivers of Investment: Output Changes versus Other Factors

9. Three types of investment models are used to explain the dynamics of private non-residential investment, following Lee and Rabanal (2010): 1) an accelerator model (Clark, 1917; and Jorgenson, 1971); 2) a neoclassical model (Jorgenson, 1971; and Caballero, 1994); and 3) a bond-market model (Philippon, 2009; Bloom, 2009; Lee and Rabanal, 2010).6Appendix 1 presents data sources and definitions.

Are output changes enough to explain the decline in investment (Accelerator Model)?

10. The first step is to explore whether changes in the output are sufficient to explain investment dynamics in the euro area. The accelerator model relates real investment to past changes in real output, taken to be the primary determinants of the changes in the desired capital stock. A common approach is to run these regressions on the investment-to-capital stock ratio:

where I is real private non-residential investment, K is the total capital stock, ΔY is the change in real GDP. 7

11. The results indicate that changes in output can capture broad changes in the investment but cannot fully account for the decline in investment after the crisis in most cases (Appendix 2). Lags of changes in the real GDP (up to 12) are correctly signed and significant. While the model provides a good fit for overall trends, real non-residential private investment, particularly during the second phase of the crisis, is lower than in-sample fitted values, with the exception of Spain. The model seems to track Spain’s investment closely, implying that output has played an important role in investment dynamics. For most countries, underinvestment becomes smaller towards the end of the sample. The model does not seem to explain well the behavior of total investment in Greece and Ireland. As a robustness check, the regressions are run for machinery and equipment investment in Ireland and Germany (with data up to 2013Q4). For both cases, the results are broadly the same.

Euro Area




Do cost of capital and financial constraints matter for investment (neoclassical model)?

12. Since output developments cannot explain fully the decline in investment after the crisis, we explore whether the cost of capital and financial constraints are additional impediments. In the neo-classical model, current investment is a function of the lags of changes in desired capital stock, which in turn is determined by the cost of capital. Under the additional assumption that the cost of capital is equal to its marginal product, investment can be related to the past changes in output and changes in the real cost of capital. This baseline specification is then augmented with a variable to capture credit rationing (based on a question on financial constraints from the EC consumer and business survey). Up to 12 lags are found to be significant.

13. Both nominal and real costs of capital are elevated for the stressed countries (See Appendix 1 for definition of the cost of capital). While reduced policy rates have translated into lower cost of borrowing in the core countries, these rates have remained elevated in the stressed countries—a sign of continued fragmentation.

Nominal Cost of Capital, percentage points

Real Cost of Capital, percentage points

14. The high cost of capital and financial constraints help explain part of the decline in investment in Italy, but actual investment remains below the estimated level for others (Appendix 3).8 Lagged coefficients of real cost of capital are generally positive and statistically significant. This is consistent with the expectation that lagged desired investment—captured by the changes of the ratio of output to the cost of capital—has a positive effect on investment in the current period. In addition, in line with expectations, the financial constraint variable is significant for investment in the euro area, Spain, and Italy. The gap between the actual and fitted investment in the euro area and Italy closes towards the end of the estimation period. Focusing on core countries and including longer lags for Germany and France, the underinvestment result survives for Germany but not for France.9 The robustness checks using series for machinery and equipment investment in Ireland and Germany produce marginally better results, but the broad messages are the same.

Euro Area




15. Out-of-sample projections also imply underinvestment. For all countries, one-step-ahead forecasts from 2008Q3 onward produce projected investment levels that are above realized investment levels, particularly during the second phase of the crisis. In the case of Germany, during the first phase of the crisis, the decline in projected investment was deeper than the actual decline. This has reversed during the second phase of the crisis. To test whether the crisis has changed the investment dynamics, an intercept and interaction dummies are added to the specification. While the intercept terms are generally significant, the results are mixed for the interaction terms (All results are available upon request).

Do other factors (uncertainty, leverage, and cash flow) play a role in investment dynamics?

16. To account for other factors that could potentially weigh on investment, we use a more eclectic model. Philippon (2009) suggests using bond prices instead of equity prices to estimate the value of Tobin’s Q. The proposed measure, called “Bond Market’s Q”, is a function of the real risk-free rate, the spread between bond yields and government bonds, leverage, and uncertainty. We substitute the real rate with the real lending rate for non-financial corporations in our baseline model, and following Lee and Rabanal (2010), include a measure of cash flow.

17. The model captures any additional impact on investment from uncertainty and corporate leverage. The ratio of private non-residential investment to total capital stock is modeled as a function of overall real lending rates to NFCs, corporate bond spreads, uncertainty, corporate leverage, and the cash flow-to-sales ratio. To account for demand effects, we augment the baseline model with changes in real output over total capital stock. This also allows us to compare this model with the accelerator and neoclassical models. Finally, similar to the neoclassical model, we control for financial constraints to account for possible credit rationing (Appendix 4):10

18. High uncertainty is associated with low investment, particularly for the stressed countries. In the baseline model (without controlling for output changes and financial constraint variable), uncertainty reduces investment in most countries and in the euro area as a whole, though the effect is fairly small. An increase in uncertainty by one standard deviation reduces the investment-to-capital-stock ratio by around 0.1 percentage points—except for Greece and Ireland for which the effect is larger). The results remain broadly unchanged when we control for output changes and financial constraints.

19. Higher corporate leverage is associated with weak investment in some countries. In the baseline model, corporate leverage reduces investment in Italy, Portugal, France, and Spain by between 0.1 and 0.4 percentage points for every 10 percentage-point increase in the debt-to-equity ratio. Controlling for output changes and financial constraints, leverage is still important for Italy, Portugal, and France. Cash flow is significant for a few countries in the baseline model, but only has the expected positive sign for Germany, Greece, and, after controlling for the changes in output, Spain.

20. Corporate bond spreads and real lending rates are significant and correctly signed only for a few countries. The former are significant for Ireland, as well as for Germany and Spain once we control for output changes and financial constraints. A 100 basis-point increase in the spread of corporate over government bond yields decreases investment by about 0.1-0.8 percentage points. Real lending rates (deflated by GDP deflator) are correctly signed and significant for determining investment in Italy once we control for the output changes and financial constraints.11 The financial constraint variable is significant for Italy and Portugal.

21. Overall, the model seems to work better for stressed countries, in particular for Spain and Italy, and to a lesser extent for Portugal and the euro area as a whole. It performs comparatively poorly for Germany and France.12 The robustness checks using the series for machinery and equipment investment in Ireland and Germany (with data up to 2013Q4) produced broadly similar results.

Euro Area


The magnitude of missing investment

22. Since the European debt crisis, investment has been systematically lower than its estimated level, except for Spain. To better gauge how much investment has been missing since the start of the European sovereign crisis, we look at the cumulative underinvestment since then. Overall, controlling only for output, the cumulative underinvestment is about 3–6 percent of GDP (excluding Spain). Once we also control for other determinants, the cumulative underinvestment declines substantially to about ½–2 percent. In Spain, output changes alone are enough to explain much of the decline in investment. However, other factors such as cost of capital, financial constraints, and uncertainty also important factors affecting investment in Spain, implying that these factors may affect investment through their impact on output.

Cumulative Underinvestment

(Since 2010Q2, in percent of GDP)

Notes: Germany ends in 2012Q4 and Portugal in 2013Q3.

C. Conclusions

23. Investment has been weak across the euro area. Empirical evidence suggests that output dynamics can explain the broad trends in investment, including its collapse after the financial crisis. In particular, output accounts for the behavior of investment in Spain. In other countries (including in Germany and France), private non-residential investment has been lower than implied by output developments only since the onset of the crisis.

24. In addition to output dynamics, there is evidence that the cost of capital and financial constraints matter for investment. Real cost of capital and financial constraints have been important factors in constraining investment, particularly for Italy and Spain. Nevertheless, controlling for these factors reduces the underinvestment only for Italy.

25. High uncertainty and corporate sector leverage are additional impediments to investment, particularly in Italy, Spain, Portugal, and France. After controlling for all of these factors, investment (in cumulative terms) is lower than its estimated level by up to about 2 percent of GDP since the beginning of the European debt crisis.

26. Investment is expected to pick up as the recovery strengthens and uncertainty declines. However, a sustained recovery in investment will require dealing with the corporate debt overhang and financial fragmentation. Corporate debt-to-equity remains elevated in some stressed countries, and the deleveraging process is still at an early stage. At the same time, borrowing costs need to be substantially lower particularly for smaller firms.

27. Future work will focus on firm level investment, particularly for SMEs. Firm-level analysis will supplement macro-level regressions. The use of microeconomic data will allow to differentiate between the investment patterns of large and small corporations, as well as the impact of firm-specific variables, such as cash-flow, leverage, and Tobin’s Q.

Appendix 1. Data

1. Real investment. Investment data are downloaded from the Eurostat. Private non-residential investment is the sum of investment in transport and other machinery and equipment, cultivated assets and intangible fixed assets.

2. Capital stock series are from AMECO database—the annual series were linearly interpolated so that the stock of capital in the last quarter would match the corresponding annual figure. Alternative measures of capital stock are also calculated using perpetual inventory method. The initial capital stock values from the AMECO capital stock were scaled by applying appropriate investment subcomponent ratios. Depreciation rates are assumed constant and equal to average rates implied by the AMECO series.

3. Real GDP on quarterly basis was obtained from the World Economic Outlook database.

4. Real cost of capital. The correct measure of the cost of capital depends on the structure of financing of the firm. The flow of funds data suggests that liabilities of non-financial corporations consist primarily of loans and equity with the share of bond financing being less than 10 percent in most periods and countries. The following formula is used for the real cost of capital:

where Di,t, Bi,t and Ei,t, are the amounts of bank loans, bonds (securities other than shares), and equity in the liabilities of non-financial corporations. For the nominal costs of different kind of capital we use MFI lending rates in a given country for new business at all maturities, li,t, for bank loan liabilities, yield on the euro area wide corporate bond index, it, for bond liabilities and the yield on 10 year government bond ci,t to price equity liabilities.13 In line with the literature, from the nominal rate, we subtract the year-on-year change in investment deflator πi,t, add the depreciation rate, which is assumed to be constant but different across countries δi and multiply the result by the relative price of investment goods (investment deflator) and output Pi,t1/pi,t. We also report the “nominal cost of capital”, which is simply the first three terms in the above equation. In addition, we also use a measure of real cost of capital for debt financing, composed of bond and bank lending (available upon request).

5. In most countries the real cost of capital has been declining throughout the 2000s, however, after the crisis Southern European countries diverged from France and Germany. Figure 1 shows the nominal and real cost of capital for the countries considered. As of the latest available data, the lowest real cost of capital is in Germany (5 percent), while Portugal has the highest cost (12.0 percent). The volatility of real cost of capital in Greece (for which only a shorter sample is available) is driven by the volatility of the investment deflator.

Figure 1.Cost of Capital Calculations

Source: Haver Analytics; and IMF staff estimates.

6. Financial constraints: This variable is from European Commission Business and Consumer Survey (quarterly). Seasonally adjusted series are for survey of manufacturing industry: percent of correspondents listing financial constraints as the factor limiting production.

7. Corporate bond prices. We use the average spread of corporate over government bonds with 1 to 5 years maturity for the euro area as a whole for all countries in the sample, to proxy corporate bond market conditions. This measure inherently gives more weight to large euro area economies and applies to large firms. (Merill-Lynch indices, Bloomberg). This is in basis points.

8. Uncertainty index: Bloom (2009). Natural log of uncertainty index*100

9. Corporate sector leverage: Debt-to-equity ratio from the ECB (in percent).

10. Cash flow-to-sales: Worldscope. Corporate vulnerability unit (median).

11. Crisis dummy: crisis =1 from 2008Q3 (used only for robustness checks).

Appendix 2. Results of the Accelerator Model1
Table.Accelerator Model - Total Investments (Newey-West HAC standard error estimates)
Euro AreaGermanyFranceItalySpainGreeceIrelandPortugal
β10.32 ***0.21 ***0.26 ***0.47 ***0.41 ***0.38 **0.75 **0.29 ***
β20.21 **0.23 ***0.23 ***0.39 ***0.49 ***0.70 ***0.99 **0.26 **
β30.25 ***0.25 ***0.23 ***0.37 ***0.23 ***1.10 ***0.92 **0.26 ***
β40.22 ***0.18 ***0.20 ***0.19 **−0.050.71 ***0.76 **0.24 ***
β50.13 *0.13 ***0.20 ***0.27 ***0.12 *1.04 ***0.370.21 **
β60.16 ***0.10 **0.17 ***0.18 ***0.39 ***1.12 ***0.300.19 **
β70.17 **0.070.09 *0.28 ***0.090.82 ***0.280.22 ***
β80.060.09 *0.12 **0.24 ***0.000.57 **0.10
β90.10 **0.12 **0.11 ** **
β100.09 *0.070.10 **0.18 ***0.15 ***0.71 *0.16 *
β110.050.10 *0.080.26 ***0.06 *1.04 ***0.17 *
β120.18 ***0.20 ***0.38 ***0.75 **0.16 *
δ3.43 ***5.99 ***3.98 ***4.29 ***2.76 ***10.65 ***8.92 ***3.81 ***
Akaike info criterion0.50−0.880.330.350.990.690.500.80
Schwarz Bayesian criterion -1.31−0.48−1.57−0.05−2.211.334.400.11
S.E. of regression0.

Figure 1Accelerator Model: Private Non-residential Investment/Capital Ratio

Sources: Eurostat; IMF. World Economic Outlook database; OECD, Analytical database; European Commission, AMICO database; and IMF staff calculations.

Note: Total investment for Greece and Ireland.

Appendix 3. Results of the Neoclassical Model
Table 2:Neoclassical Model augmented with Financial Constraints: Estimates with Newey West Standard Errors
Euro AreaGermanySpainFranceGreeceIrelandItalyPortugal
Adjusted R-squared0.6730.5280.8240.2170.8500.4560.8570.728
D-W Statistic0.3800.3760.4880.1801.0170.1120.6020.519
Appendix 4. Results of the Bond Market Model
Table 2.Bond Market Model (Controlling for Output Changes and Financial Constraints)
Euro AreaGermanySpainFranceGreeceIrelandItalyPortugal
β160.112 (0.08)0.408 (0.339)0.107 (0.067)0.214*** (0.065)
β170.1* (0.058)0.131** (0.06)
DW statistic0.711.261.550.51.571.091.711.19

Figure 1.Bond Market Model

(Ctrolling for Output Changes and Financial Constraints)

Sources: AMECO database; Haver Analytics; Eurostat; and IMF staff estimates.


Prepared by Bergljot Barkbu, Pelin Berkmen, Pavel Lukyantsau, Sergejs Saksonovs, and Hanni Schoelermann. We are grateful for the comments provided by Philip Vermeulen and other participants at the ECB seminar, as well as the European Commission staff.

Investment in percent of GDP remains below its pre-crisis long-term (1995–2007) level, particularly for stressed countries.

For these charts and regressions, private non-residential investment data are obtained from Eurostat to ensure consistency and comparability across countries. Looking into other data sources also shows weak investment dynamics. For example, real fixed-investment in equipment in Germany and real investment by NFCs in France, equipment and transportation machinery in the euro area are also weaker than their pre-crisis levels.

In addition to these standard factors, investment in many smaller European countries has been affected by availability of the EU structural funds. For instance, there are indications that investment in Portugal in certain sectors was too high prior to the crisis (see OECD WP 994).

Select countries are Germany, France, Italy, Spain, Portugal, Ireland and Greece. We use quarterly data for 1990/99 and 2012/13.

A model based on Tobin’s Q was also estimated. This model relates the real investment-to-capital ratio to the ratio of firm value to the replacement cost of the existing capital stock (Tobin, 1969; and Hayashi, 1982). Alternative definitions of Tobin’s Q (for NFCs) are used: i) interpolated from annual Tobin’s Q (CVU, Woldscope); ii) price-to-book ratio; iii) stock prices deflated by GDP deflator. The model also controls for firms’ leverage (debt-to-asset ratio) and cash-flow (CVU, Worldscope). Among the Tobin’s Q proxies used, only price-to-book ratio appears to be significant in a few specifications for Germany, France, and Portugal. Controlling for endogeneity (by two-stage least squares) and running the regressions for the pre-crisis period, the significance of the results increases: price-to-book ratio and leverage are significant and correctly signed for Germany, Greece, Portugal, and the euro area. Overall, however, model performance remains weak.

For Ireland and Greece, total real investment is used. As the residuals are highly correlated—a common result in the literature—we report Newey-West standard errors. Note that the constant δ can be interpreted as an indirect estimate of the depreciation rate.

Similar to the accelerator model, the residuals are serially correlated and the model is not a good fit for Greece and Ireland. Macroeconomic adjustment in those countries has been especially severe, which, combined with the relatively small sample, may affect estimation results. In addition, the sample size is smaller for Greece, limited by cost of capital calculations. Appendix 3 presents the results for the extended model; the results for the baseline model without financial constraints are available upon request.

We have also modeled the short and long-run investment dynamics using a vector error correction model, controlling for cost of capital, output, and labor costs. While we did not obtain consistently significant coefficients for the cost of capital measure, output and labor costs were significant, showing that higher labor costs dampen investment growth. Augmenting the VECM model with indicators of capacity utilization and uncertainty generally failed to establish significant results.

Appendix 4 presents the results for the extended model. The results for the baseline and other steps are available upon request.

The coefficients for the euro area, Germany, and Spain have the reverse sign, which is a common finding in the literature, possibly reflecting difficulties in identifying credit demand and supply.

Pérez Ruiz (2014) uses a broader set of determinants to explain the level of business investment in France. The model provides a good fit for France.

We have experimented with alternative approaches to price equity capital – such as variations on the dividend growth model; however, they tend to produce counterintuitive implications for the ranking of the cost of capital across different countries. Using a 10 year government bond establishes a sensible lower bound for the cost of equity and, assuming that the risk premium is constant, is not expected to affect the results. For the euro area we use the simple average of the 10-year bond yields in France, Germany, Spain and Italy.

Results are in percent terms. Total investment is used for Greece and Ireland.

Other Resources Citing This Publication