compared to the Basel Credit Gap. However, when taking the best performing HP-filter specification’s AUC value, these differences are no longer statistically significant.
When looking at the graph of the credit gap estimates of the model for the United Kingdom, the VEC specification seems to be a reasonable fit for the domestic credit cycle.
Importantly, it signals substantial positive imbalances in the lead-up to the GFC, more so than the BCG. This is also true for France, where the model with house prices indicates a positive gap for most of the late 1990s and the period before the GFC. Sweden’s house price VEC specification is more volatile than the Netherlands’ specification until the early 90s. After that, it signals more strongly building financial imbalances in the lead-up to the GFC than the BCG.
For Germany, however, while the AUC-values would indicate otherwise, the VEC model with equity prices seems a poor fit. It does signal building imbalances correctly for the 90s, but indicates positive imbalances during the SDC, which seems highly unlikely. The HP-filter does the opposite. For the United States, the model with house prices seems to move quite puzzlingly. Importantly, it fails to signal building imbalances for the financial crisis in the 80s and the GFC. While it did drop as a result of the deflation of the house price bubble in 2008, it does not signal that positive financial vulnerabilities were prevalent before.
Thus, while the structural VEC model with asset prices performs strongly for the Netherlands, these results do not hold up as strongly in an international context. This has three implications. First, it underlines the importance of a detailed analysis focused on one nation at a time, as has been done in this study. As mentioned, I used the specifications tailored for the Netherlands, and ran them for these countries. Perhaps with more fine-tuning of the number of lags, cointegration relations, and specific variables, the results for a structural VEC model would improve for these other countries. Second, these results highlight the heterogeneity of domestic financial cycles (Aldasoro et al., 2020). If the domestic financial cycles and the drivers behind them would be similar to each other, then the results for the Netherland would also be reflected in the results for the other countries. Third, and interrelated to the second point, the results underline the importance of national discretion in macroprudential policy-making. Considering the substantial differences in both the optimal VEC specification (some with the composite indicator, others with house prices) and the time periods that the credit gaps turn positive, macroprudential policy should be decided by national policymakers who have the firmest grasp on what the situation regarding financial stability is in their country.
al. 2012; Drehmann & Yetman, 2018, 2020). But the BCG has several weaknesses. In particular, due to the inertia of its long-run trend, the BCG tends to turn persistently negative after a pronounced credit boom phase deflates. A good example of this, is the Netherlands.
This would implausibly indicate that the credit-to-GDP ratio should return to its peak from the boom phase. Importantly, this can hamper the ability of the BCG to signal building vulnerabilities during the next boom phase of the credit cycle, especially when credit recovers quickly. This is not an issue particular to the BCG, but a flaw in all purely statistical indicators.
Indicators based on structural models, which link credit developments to underlying fundamentals, might provide more accurate signals of credit imbalances, and as such give more helpful information to macroprudential policymakers on the activation of the CCyB and other cyclical macroprudential measures.
In this paper, I assess whether a structural Vector Error Correction Model (VECM) with asset prices improves on the Basel Credit Gap in terms of the credit gap estimation for the Netherlands. I use a relatively long sample of quarterly data from 1970 to 2020, which allows for the in-depth analysis of several boom-and-bust phases of the Dutch credit cycle. There are three particularly remarkable components of this study with respect to the previous literature. First, I estimate the model in real-time. This is of particular relevance for policymakers as they make their decisions in real-time as well. Furthermore, this allows for a truly proper comparison with the BCG, as it is also estimated in real-time. Second, I qualitatively analyse the model in depth, as opposed to only assessing its predictive properties for financial crises. This allows for the assessment of the indicator across the entire length of the credit cycle, rather than right before its bust. Third, I scrutinise whether adjusting the underlying smoothing parameter of the BCG has an effect on the indicator’s estimated credit gaps. Very few studies that propose new indicators do this, while evidence suggests that this might significantly improve its performance as an indicator (Galan, 2019).
My results suggest that a structural VEC model that includes a composite indicator with house- and equity prices and the real interest rate improves on the BCG in terms of the credit gap estimation for the Netherlands. Similar to the BCG, credit gaps turn positive ahead of financial crises, but unlike the BCG, the credit gaps do not stay negative for an extended period of time after a prolonged credit boom. This more accurately fits prevailing financial market conditions, and suggests that this measure is better at signalling building financial vulnerabilities across the entire credit cycle. What is more, the predictive properties for financial crises of the VEC model with the composite indicator for asset prices and the real interest rate are significantly better than the BCG’s. This is interesting, as the BCG was chosen for its strong predictive properties of financial crises (BCBS, 2010; Drehmann et al., 2012;
Drehmann & Yetman, 2020). These results hold up also when running several robustness checks for the Netherlands, and other types of specifications.
Both the statistical and qualitative robustness of the results also apply to the main specification of the VECM when credit from banks is used, rather than credit from all sectors.
As such, the VECM can add a layer of depth to the information that policymakers are able to extract from the model. Considering that the CCyB is targeted at banks, I consider this of added value for policymakers. Separating credit by type of borrower, however, leads to highly unstable results. Results from the tested specifications also suggests that it is the combination of the upward movement of equity and house prices in tandem with periods of credit growth when credit growth in the Netherlands is most likely to be excessive and financial systemic risk is building up. This is in line with the findings of Reinhart and Rogoff (2012), Schüler et al.
(2018) and, more recently, Greenwood et al. (2020). However, it is at odds with some of the
financial cycle literature that states that equity prices contribute to more ‘noise’ and can be a distraction in characterising the financial cycle (e.g., Drehmann et al., 2012; Borio, 2014).
The inclusion of the real interest rate seems to convey an important informational component to the credit gap estimation, as identified by its highly significant coefficients in most specifications, and the higher statistical robustness of the credit gaps that included real interest rates. This fits with much of the literature on the link between real interest rates and credit (e.g., Bernanke & Gertler 1987; Kiyotaki & Moore 1997; Jermann & Quadrini 2010;
Bianchi & Mendoza 2018; Baba et al., 2020).
The robustness of the model’s results for the Netherlands are not as strong in an international context when running several specifications for France, Germany, Sweden, the United Kingdom, and the United States. This has three implications. First, it underlines the importance of a detailed analysis focused on one nation at a time, as has been done in this study. As mentioned, I used the specifications tailored for the Netherlands, and ran them for these countries. With more fine-tuning of the number of lags, cointegration relations, and specific variables, the results for a structural VEC model could improve for these other countries. Second, these results highlight the heterogeneity of domestic financial cycles (SOURCES?). If the domestic financial cycles would be similar to one another, and the drivers behind these financial cycles, then the results for the Netherland would also be reflected in the results for the other countries. Third, and interrelated to the second point, the results underline the importance of national discretion in macroprudential policymaking. The substantial differences in both the optimal VEC specification (some with the composite indicator, others with house prices) and the time periods that the credit gaps turn positive, suggests that macroprudential policy should be tailored to nations rather than blocs.
While my findings suggest that my model is a significant improvement on the credit gap estimation as compared to the BCG for the Netherlands, it is an addition to a list of many other multivariate indicators that seem to perform better than the BCG (e.g., Schüler et al., 2017; Lang & Welz, 2019; Baba et al, 2020; Galan & Mencia, 2021). Therefore, one path of future research could be to run a ‘horse-race’ consisting of a large range of multivariate financial vulnerabilities indicators. Drehmann and Yetman (2018, 2020) estimated a range of univariate indicators in real-time and compared their early warning properties for systemic crises, and found the BCG to be the best indicator overall, but this has not been done for multivariate models. Running a ‘horse-race’ between the multivariate indicators, to determine which indicator performs best in various contexts, can help policymakers in determining which indicator to use, and might offer a new ‘common reference point’ as suggested by the BCBS.
Appendix I
This lists all the tested specifications by their corresponding number. For every specification it includes the variables specified, the trend assumption, whether the asset prices variable is in levels or logs, the number of lags in the short-run equation, the AUC-value for the
window of 5 to 16 quarters ahead of a financial crisis, and the number of times the model did not converge.
Main Specifications Credit from All Sectors
Specifi-cation # Variables Trend
Assumption Level or Log
# of
Lags
AUC-value #
Nonconvergence 1 Credit, GDP**, Asset prices***,
Real Interest Rate Constant Log 5 0,971 1
1.1 Credit*, GDP**, Asset prices***,
Real Interest Rate Constant Log 2 0,733 0
1.2 Credit, GDP, Asset prices,
Nominal Interest Rate Constant Log 5 0,832 16
2 Credit, GDP, House Prices, Real
Interest Rate Constant Log 5 0,544 5
2.1 Credit, GDP, House Prices, Real
Interest Rate Constant Log 4 0,48 3
2.2 Credit, GDP, House Prices,
Nominal Interest Rate Constant Log 5 0,557 20
3 Credit, GDP, House Prices,
Equity Prices, Real Interest Rate Constant Log 5 0,854 7 3.A Credit, GDP, Equity Prices, Real
Interest Rate Constant Log 6 0,841 8
3.A2 Credit, GDP, Equity Prices,
Nominal Interest Rate Constant Log 6 0,826 17
Main Specification Credit from All Sectors, without Real Interest Rates
4 Credit, GDP, Asset Prices Constant Log 5 0,642 0
5 Credit, GDP, House Prices Constant Log 5 0,558 34
5,1 Credit, GDP, House Prices Constant Log 4 0,535 10 6 Credit, GDP, House Prices,
Equity Prices Constant Log 5 0,241 21
Main Specifications with credit from Banks only
Specifi-cation # Variables Trend
Assumption Level or Log
# of
Lags
AUC-value #
Nonconvergence 7 Bank Credit, GDP, Asset Prices,
Real Interest Rate Constant Log 5 0,951 1
8 Bank Credit, GDP, House Prices,
Real Interest Rate Constant Log 5 0,718 11
9 Bank Credit, House Prices,
Equity Prices, Real Interest Rate Constant Log 5 0,823 3 Main specifications with credit from banks only, without real interest rates
10 Bank Credit, GDP, Asset Prices Constant Log 5 0,637 1
11 Bank Credit, GDP, House Prices Constant Log 5 0,418 50
12 Bank Credit, GDP, House Prices,
Equity Prices Constant Log 5 0,648 10
Other Specifications (same trend)
Household Credit (from all sectors)
13 Household Credit, GDP, Asset
Prices, Real Interest Rate Constant Log 5 78
13,1 Household Credit, GDP, Asset
Prices, Real Interest Rate Constant Log 2 78
Specifi-cation # Variables Trend
Assumption Level Log or
# of
Lags
AUC-value #
Nonconvergence 14 Household Credit, GDP, House
Prices, Real Interest Rate Constant Log 5 46
15 Household Credit, GDP, House Prices, Equity Prices, Real
Interest Rate
Constant Log 5 79
Household Credit without Real Interest Rates
16 Household Credit, GDP, Asset
Prices Constant Log 5 70
17 Household Credit, GDP, House
Prices Constant Log 5 69
18 Household Credit, GDP, House
Prices, Equity Prices Constant Log 5 69
Company Credit 19 Company Credit, GDP, Asset
Prices, Real Interest Rate Constant Log 5 61
19,1 Company Credit, GDP, Asset
Prices, Real Interest Rate Constant Log 2 61
20 Company Credit, GDP, House
Prices, Real Interest Rate Constant Log 5 47
21 Company Credit, GDP, House Prices, Equity Prices, Real
Interest Rate
Constant Log 5 60
Company Credit without Real Interest Rates
22 Company Credit, GDP, Asset
Prices Constant Log 5 59
23 Company Credit, GDP, House
Prices Constant Log 5 50
24 Company Credit, GDP, House
Prices, Equity Prices Constant Log 5 78
Level Specifications Credit from All Sectors (Levels)
Specifi-cation # Variables Trend
Assumption Level or Log
# of
Lags
AUC-value #
Nonconvergence 25 Credit, GDP, Asset prices, Real
Interest Rate Constant Level 5 0,878 2
26 Credit, GDP, House Prices, Real
Interest Rate Constant Level 5 0,875 27
26,1 Credit, GDP, House Prices, Real
Interest Rate Constant Level 4 0,529 3
27 Credit, GDP, House Prices,
Equity Prices, Real Interest Rate Constant Level 5 0,923 13
Credit from All Sectors (Levels) without Real Interest Rates
28 Credit, GDP, Asset prices Constant Level 5 0,646 1
29 Credit, GDP, House Prices Constant Level 5 0,628 40
30 Credit, GDP, House Prices,
Equity Prices Constant Level 5 1 26
Credit from Banks (levels) 31 Bank Credit, GDP, Asset prices,
Real Interest Rate Constant Level 5 0,974 2
32 Bank Credit, GDP, House Prices,
Real Interest Rate Constant Level 5 0,802 5
33 Bank Credit, GDP, House Prices,
Equity Prices, Real Interest Rate Constant Level 5 0,75 13
Credit from Banks (levels) without Real Interest Rates
34 Bank Credit, GDP, Asset prices Constant Level 5 0,509 3 35 Bank Credit, GDP, House Prices Constant Level 5 0,543 45
Specifi-cation # Variables Trend
Assumption Level Log or
# of
Lags
AUC-value #
Nonconvergence 36 Bank Credit, GDP, House Prices,
Equity Prices Constant Level 5 0,938 15
Credit from all sectors, in Nominal Terms 37 Nominal Credit, Nominal GDP,
Nominal Asset Prices, Net Interest Rates
Constant Log 5 0,616 65
38 Nominal Credit, Nominal GDP, Nominal House Prices, Nominal
Interest Rates
Constant Log 5 0,794 33
39 Nominal Credit, Nominal GDP, Nominal House Prices, Nominal
Equity Prices, Nominal Interest Rates
Constant Log 5 0,619 35
Credit from all sectors, in Nominal Terms without Interest Rates 40 Nominal Credit, Nominal GDP,
Nominal Asset Prices Constant Log 5 0,471 62
41 Nominal Credit, Nominal GDP,
Nominal House Prices Constant Log 5 0,741 23
42 Nominal Credit, Nominal GDP, Nominal House Prices, Nominal
Equity Prices
Constant Log 5 0,538 12
Bank Credit, all in Nominal Terms 43 Nominal Credit, Nominal GDP,
Nominal Asset Prices, Net Interest Rates
Constant Log 5 0,997 26
44 Nominal Credit, Nominal GDP, Nominal House Prices, Nominal
Interest Rates
Constant Log 5 0,756 29
45 Nominal Credit, Nominal GDP, Nominal House Prices, Nominal
Equity Prices, Nominal Interest Rates
Constant Log 5 0,561 35
Bank Credit in Nominal Terms without Interest Rates 46 Nominal Credit, Nominal GDP,
Nominal Asset Prices Constant Log 5 0,468 36
Specifi-cation # Variables Trend
Assumption Level Log or
# of
Lags
AUC-values #
Nonconvergence 47 Nominal Credit, Nominal GDP,
Nominal House Prices Constant Log 5 0,91 27
48 Nominal Credit, Nominal GDP, Nominal House Prices, Nominal
Equity Prices
Constant Log 5 56
Different Trend Assumptions Credit from all sectors
49 Credit, GDP, Asset Prices, Real
Interest Rate Trend Log 5 0,872 26
50 Credit, GDP, Asset Prices, Real
Interest Rate Restricted
Trend Log 5 0,948 28
1 Credit, GDP, Asset Prices, Real
Interest Rate Constant Log 5 0,971 1
51 Credit, GDP, Asset Prices, Real
Interest Rate Restricted
Constant Log 5 0,971 12
52 Credit, GDP, Asset Prices, Real
Interest Rate None Log 5 0,439 7
Credit from all sectors, without Real Interest Rates
53 Credit, GDP, Asset Prices Trend Log 5 0,701 15
54 Credit, GDP, Asset Prices Restricted
Trend Log 5 0,705 35
4 Credit, GDP, Asset Prices Constant Log 5 0,642 0
55 Credit, GDP, Asset Prices Restricted
Constant Log 5 0,701 15
56 Credit, GDP, Asset Prices None Log 5 0,701 15
Credit from all sectors, with House Prices 57 Credit, GDP, House Prices, Real
Interest Rate Trend Log 5 0,845 65
Specifi-cation # Variables Trend
Assumption Level Log or
# of
Lags
AUC-value #
Nonconvergence 58 Credit, GDP, House Prices, Real
Interest Rate Restricted
Trend Log 5 0,668 60
2 Credit, GDP, House Prices, Real
Interest Rate Constant Log 5 0,544 5
59 Credit, GDP, House Prices, Real
Interest Rate Restricted
Constant Log 5 0,779 89
60 Credit, GDP, House Prices, Real
Interest Rate None Log 5 0,779 89
Bank Credit
61 Bank Credit, GDP, Asset Prices,
Real Interest Rate Trend Log 5 0,863 31
62 Bank Credit, GDP, Asset Prices,
Real Interest Rate Restricted
Trend Log 5 0,939 35
7 Bank Credit, GDP, Asset Prices,
Real Interest Rate Constant Log 5 0,951 1
63 Bank Credit, GDP, Asset Prices,
Real Interest Rate Restricted
Constant Log 5 0,974 11
64 Bank Credit, GDP, Asset Prices,
Real Interest Rate None Log 5 0,365 8
Bank Credit without Real Interest Rates
65 Bank Credit, GDP, Asset Prices Trend Log 5 0,613 8
66 Bank Credit, GDP, Asset Prices Restricted
Trend Log 5 0,432 9
10 Bank Credit, GDP, Asset Prices Constant Log 5 0,637 1
67 Bank Credit, GDP, Asset Prices Restricted
Constant Log 5 0,433 8
68 Bank Credit, GDP, Asset Prices None Log 5 0,797 0
Bank Credit, with House Prices
Table 8. All tested specifications listing the number of the specification, the included variables, the trend assumption, whether the asset price variable was specified in logs or levels, the number of lags in the short-run equation, the AUC-value for the 5 to 16 quarters ahead window, and the times the model did not converge over the recursive estimates.
Source: author’s Own calculations.
Specifi-cation # Variables Trend
Assumption Level Log or
# of
Lags
AUC-value #
Nonconvergence 69 Bank Credit, GDP, House Prices,
Real Interest Rate Trend Log 5 0,779 3
70 Bank Credit, GDP, House Prices,
Real Interest Rate Restricted
Trend Log 5 0,747 1
8 Bank Credit, GDP, House Prices,
Real Interest Rate Constant Log 5 0,718 11
71 Bank Credit, GDP, House Prices,
Real Interest Rate Restricted
Constant Log 5 0,895 40
72 Bank Credit, GDP, House Prices,
Real Interest Rate None Log 5 0,781 61
Appendix II
These include figures on a selection of coefficient estimates, and their corresponding z-scores.
Figure 15. Coefficients of the recursive estimates of the adjustment parameters for the asset price variable for Models 1, 7, and 31. Source: author’s own calculations.
Figure 16. Z-scores for coefficients of the recursive estimates of the adjustment parameters for the asset price variable for Models 1, 7, and 31. Source: author’s own calculations.
-30 -20 -10 0 10 20 30
-0,2 -0,15 -0,1 -0,05 0 0,05 0,1 0,15 0,2
1980q 1
1982q 3
1985q 1
1987q 3
1990q 1
1992q 3
1995q 1
1997q 3
2000q 1
2002q 3
2005q 1
2007q 3
2010q 1
2012q 3
2015q 1
2017q 3
2020q 1
Alpha Estimates Asset Prices of Composite Indicator Models
Model 1 Model 7 Model 31
-3 -2 -1 0 1 2 3 4 5
1980q 1
1982q 3
1985q 1
1987q 3
1990q 1
1992q 3
1995q 1
1997q 3
2000q 1
2002q 3
2005q 1
2007q 3
2010q 1
2012q 3
2015q 1
2017q 3
2020q 1
Z-scores Alpha Estimates Asset Prices
Model 1 Model 7 Model 31
Figure 17. Coefficients of the recursive estimates of the adjustment parameters for the real interest rate variable for Models 1, 7, and 31. Source: author’s own calculations.
Figure 18. Z-scores for coefficients of the recursive estimates of the adjustment parameters for the real interest rate variable for Models 1, 7, and 31. Source: author’s own calculations.
-0,5 0 0,5 1 1,5 2 2,5 3 3,5 4
1980q 1
1982q 3
1985q 1
1987q 3
1990q 1
1992q 3
1995q 1
1997q 3
2000q 1
2002q 3
2005q 1
2007q 3
2010q 1
2012q 3
2015q 1
2017q 3
2020q 1
Alpha Estimates Real Interest Rates of Composite Indicator Models
-1,5 -1 -0,5 0 0,5 1 1,5 2 2,5 3
1980q 1
1982q 3
1985q 1
1987q 3
1990q 1
1992q 3
1995q 1
1997q 3
2000q 1
2002q 3
2005q 1
2007q 3
2010q 1
2012q 3
2015q 1
2017q 3
2020q 1
Z-scores of Alpha Estimates Real Interest Rate
Model 1 Model 7 Model 31
Appendix III
Model
Number Type Asset
Prices Type Credit Maximum rank (trace
statistic)1
LR test of identifying Restrictions2
Modulus - eigenvalue stability Condition3
1 Composite All 1 (24,41) 0,836 0,95217
2 House All 1 (23,0475 0,079 0,96427
3.A Equity All 1 (25,06) 0,465 0,90177
7 Composite Bank 1 (20,10) 0,162 0,89639
31 Composite Bank 1 (20,04) 0,989 0,91964
All models have 5 lags (as indicated by the AIC), and include the real interest rate. 1 The critical value at the 5% level for the trace statistic is 29,68 for all models. 2 the p-value of the Chi-squared statistic of the long-run test of the identifying restrictions where the null hypothesis is that the imposed constraints in the long-run equation are valid. 3 A modulus of the eigenvalue that is strictly lower than one, indicates that the cointegration equation
is stable and the correct number of cointegration relations is specified.
Table 9. Key pre- and post-estimation test results of the main VEC specifications.10
10 Other results can be shared upon request.
Figure 19. Plots the AUC-values for the 5 to 16 quarters ahead of a financial crisis against the number of times the VEC specifications converged. It is zoomed in to show the most
accurate models. Statistically better fitting models will be in the top-right corner of the figure. The model numbers correspond to the specification numbers in Appendix I.
1
2 3
0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
100 120 140 160 180 200
AUC-score
Non-missing Observations
AUC-values (5-16 quarters ahead) and Times the Model Converged