Introduction
The Senior Loan Officer Opinion Survey on Bank Lending Practices (SLOOS) is one of the foremost sources of information about the supply of, and demand for, bank credit used by policymakers and market participants. However, answers to the SLOOS cannot be easily aggregated because the survey uses a qualitative, ordinal scale, and the criteria used to characterize quarterly changes in lending standards on that scale may differ across banks, loan types, direction of the change, and over time. Thus, reported easings and tightenings cannot be converted to a cardinal scale and summed across time without making strong assumptions about those criteria. As a result, the quarterly changes in lending standards reported qualitatively in the regular SLOOS cannot be used to identify an underlying level of standards, a concept that many users of the data would like to elicit.²

In order to supplement the quarterly responses with information about the level of standards, the SLOOS recently began asking once per year for the level of standards relative to a longer-run benchmark. Answers about levels and changes in standards most likely contain different information for at least two reasons. First, the range of possible responses may cause banks to answer that standards remained unchanged in a quarter unless they had changed those standards materially, whereas smaller changes in the same direction over the course of a year may cumulate to the extent that banks will report a change in levels at an annual frequency. Second, banks' willingness to reveal their true actions may differ depending on the type of question or frequency of the response. In fact, quarterly responses about changes in standards exhibit a noticeable skew toward tightening over time, whereas questions about levels suggest that they have consistently eased.³

This note shows that banks' annual answers about the level of lending standards vary over time in the same direction as aggregates of their quarterly responses about changes in lending standards over the corresponding period. This relationship is robust to certain changes in econometric models and definition of variables, but it is not clear in summary statistics. The evidence that banks' responses about levels and changes in standards are correlated suggests that banks' quarterly and annual responses to the SLOOS contain relevant information about their lending practices, while the lack of perfect correlation suggests that they each provide independent information about supply conditions in this market.⁴

Data
The data used in this note are the answers of domestic commercial banks to each quarterly SLOOS from October 2010 to July 2014, a range that spans the period in which annual questions about the levels of standards have been asked. The quarterly questions about changes in lending standards are included in all editions of the survey, while the annual questions about levels of standards are included in the July surveys from 2011 to 2014 only. The annual answers from each bank i, loan type j, and July SLOOS from year t are matched with past annual and quarterly answers from the same bank and loan type. The data include only loan types that are similarly defined in the quarterly and in the annual questions. Observations with missing answers or answers in which the bank reported that it did not originate the respective type of loan are also excluded. The final sample is composed of 907 observations of bank-loan type-year triplets.

The sample includes a large variety of banks and loan types. The data contain answers from 69 banks, which range in size from $2 billion in total assets to very large banks. The sample also includes 10 different loan types: (i) non-syndicated commercial and industrial (C&I) loans to large and middle-market firms, (ii) non-syndicated C&I loans to small firms, (iii) construction and land development loans, (iv) loans secured by nonfarm nonresidential properties, (v) loans secured by multifamily residential properties, (vi) residential real estate (RRE) loans that banks categorize as nontraditional residential mortgages, (vii) RRE loans that banks categorize as subprime residential mortgages, (viii) revolving home equity lines of credit (HELOCs), (ix) auto loans and (x) consumer loans other than credit card and auto loans.⁵

Annual questions about the levels of standards follow this general pattern:

"Using the range between the tightest and the easiest that lending standards at your bank have been between 2005 and the present, for each of the loan categories listed below, how would you describe your bank's current level of standards relative to that range? (Please respond using the following scale: 1 = near the easiest level that standards have been during this period, 2 = significantly easier than the midpoint of the range that standards have been during this period, 3 = somewhat easier than the midpoint of the range that standards have been during this period, 4 = near the midpoint of the range that standards have been during this period, 5 = somewhat tighter than the midpoint of the range that standards have been during this period, 6 = significantly tighter than the midpoint of the range that standards have been during this period, 7 = near the tightest level that standards have been during this period.)"

Changes in the level of standards are calculated as the difference between the answers of bank i about levels of standards of loan type j in the July surveys of years t and t-k, for k ∈ {1,2,3}. One feature of the survey questions is that the period of comparison increases each year to include the most recent year of data; therefore, the range between the tightest and the easiest standards from 2005 to the present may have changed over the sample period for banks that are operating near the upper or lower boundary of their range in a particular category. We account for this potential drift when interpreting variation over time in answers about levels of standards.

Quarterly questions on changes in standards follow this pattern: ⁶

"Over the past three months, how have your bank's credit standards for approving applications for loans of type X changed?

1. Eased considerably

2. Eased somewhat

3. Remained basically unchanged

4. Tightened somewhat

5. Tightened considerably"

Evidence from Summary Statistics
In Table 1, we study the correlation between answers about levels and about changes in standards by the same bank. This table shows the distribution of changes in a bank's answers about levels of standards for each loan category between the July surveys from years t-1 and t juxtaposed against the bank's answers about changes in standards for that same loan category accumulated over the four surveys from October of year t-1 to July of year t. Each cell in this table shows the number of bank-loan type-year triplets that satisfy each combination of annual changes in reported levels and quarterly changes in standards. Rows and columns are separated, respectively, by whether answers about levels and about changes indicate that banks tightened, did not change, or eased their standards.

In all three rows, the large majority of observations is concentrated in column 2, indicating that banks reported no net change in lending standards in the quarterly surveys over the previous year. In fact, most banks reported no quarterly change in standards in any particular quarter, independently of what their annual answers about levels indicate. As shown by comparing row 3 column 1 to row 3 column 3, banks that reported having eased the level of their lending standards in the annual questions were more than four times as likely to have reported easing standards, on net, over the four quarterly surveys during that comparison period. Alas, large net fractions of the banks that reported having tightened lending standards across the year covered by the annual levels questions and those reporting no change in the level also reported having eased standards, on net, in the quarterly surveys.⁷ In general, this table indicates that answers about levels and changes in standards are weakly correlated across banks and loan types over recent years.

These patterns can be explained by at least three reasons. First, banks may consider the drift in their standards during a particular quarter too small to justify, in most cases, answering that standards had changed "somewhat" rather than
"[r]emained basically unchanged." However, if the level of standards had been drifting in the same direction for the entire year, then banks may consider annual changes in standards large enough to warrant changes in the answers about levels between consecutive years. Conversely, banks that reported having changed standards during the year may feel that, despite those changes, their overall level of lending standards had not changed by enough to justify moving to a different response among the options available in the levels questions. Second, banks may prefer to answer that standards "[r]emained basically unchanged" in the quarterly surveys if they believe that this answer prevents follow-up questions from the Federal Reserve, whereas banks may not view any of the possible responses to the levels questions as being similarly "neutral" in the current environment.⁸ Third, banks that reported a level of standards that was "near the tightest (easiest) level that standards have been" in both years may have tightened (eased) further during that particular year, but the bounded responses to the levels questions would indicate no change. In any case, these frictions weaken the correlation between answers about levels and changes in standards, but do not necessarily inform us as to which of the two formats are a more accurate assessment of recent changes in lending conditions.

Econometric Evidence
For a more formal investigation of the correlation between answers about levels and changes in standards, we use the following econometric specification:

  (1)

where level_{i j t} is the level reported by bank i for loan type j in July of year t and Δ_k level_{i j 2014} ≡ level_{i j 2014} - level_{i j 2014 - k}, for k ∈ {1,2,3}. We set t = 2014 to ensure that observations do not overlap in specifications with k greater than 1. tight_{i j s} and ease_{i j s} are dummy variables that are equal to 1 if bank i reported that it tightened or eased, respectively, the standards for loan type j in the SLOOS of quarter s. S = { October t - k, January t - k + 1, …, April t, July t} is the set of SLOOS dates in the past k years. φ_i and γ_j are bank and loan type fixed effects and ε_{i j t} is an idiosyncratic shock.

α, β, and θ are parameters to be estimated. α and β should be positive and negative, respectively, because answers about levels and changes in standards should indicate changes in lending standards in the same direction. θ should be negative for two reasons. First, the reported levels of standards may revert to their means over time. Second, banks that reported levels of standards in 2014-k close to the tightest or the easiest since 2005 had more room to indicate larger changes in the opposite direction when reporting their levels in 2014. Indeed, lending standards at the beginning of the sample period were reportedly quite tight across a range of bank loan products. With subsequent gradual improvement in the economic outlook and tolerance for risk, banks reported easier levels of standards, on balance, and those changes were larger for banks that reported the tightest levels of standards at the beginning of the period.

Equation (1) uses the number of surveys in which banks reported that they tightened or eased their standards as independent variables, as opposed to dummy variables for each quarter in which banks reported that they tightened or eased standards, for three reasons. First, the dummy variables tight_{i j s} and ease_{i j s} are serially correlated, and thus using the sums and as independent variables can attenuate the problems associated with collinearity. Second, banks most often report that they did not change the standards for the respective loan type, which implies that many observations of these dummy variables are equal to zero. Indeed, for some quarters and loan types, very few banks report any changes in standards. In contrast, the sum of tightenings and easings in each category over the relevant time period are more variable, and so using those as independent variables should again lead to less collinearity and increase the precision of the coefficient estimates. Third, using the sum of recent answers to the quarterly questions reduces the number of coefficients to be estimated, which also typically leads to more precise estimates. Of course, this approach limits the ability to account for dynamics in the control variables, and thus we investigate below whether our results change if answers from different time periods enter separately in the right-hand side of equation (1).

The bank and loan type fixed effects capture the effects of important unobservable characteristics that may affect the changes over time in reported levels of standards and therefore are crucial to estimate accurately the relation between answers about levels and changes in standards. Bank fixed effects control for shocks that affect a bank's standards across the whole loan portfolio; for example, shocks to the threshold that a change in standards must exceed before the bank reports a change in standards as opposed to no change. Likewise, loan type fixed effects control for common shocks to lending standards affecting a particular loan category; for instance, the many changes in regulations for mortgages or consumer loans over the sample period.

Table 2 shows ordered probit estimates of equation (1).⁹ Column 1 shows the results using the sums of the dummies for tightening and easing from the four SLOOS from October 2013 to July 2014 and the answers about levels in July 2013 as independent variables. The coefficients of the tightening and easing variables have the expected signs--as all coefficient estimates across the four specifications in this table do--and the coefficient for easing is highly statistically significant.¹⁰ The coefficient estimates in column 1 imply that a bank that reports having eased standards in one quarter during the past year is 7 percentage points more likely to report a level of standards at least one category easier than it did in the previous July survey than a bank that reported no net change in standards during the past year.¹¹ The coefficient of the levels reported in 2013 is also statistically significant and is negative, as expected.

Columns 2 and 3 test the strength of these relationships over longer time periods. Column 2 uses the difference in the levels between July 2014 and July 2012 as the dependent variable and the sum of tightening and easing dummies since the October 2012 SLOOS and the answers about levels in the July 2012 SLOOS as independent variables. The coefficient of the tightening variable is now highly statistically significant, despite the smaller number of observations, and roughly of the same economic magnitude as the easing coefficient in column 1. Moreover, the coefficient of the easing variable is significant at the 10 percent level.

Column 3 uses the difference in the levels between July 2014 and July 2011 as the dependent variable and the sums of the quarterly dummies since the October 2011 and the answers about levels from July 2011 as independent variables. Now, neither of the coefficients on the variables representing the sum of reported changes is significant individually, likely in part because the number of observations is reduced even further. However, as shown in the memo item, the two coefficients are jointly significant at the 5 percent significance level and their magnitudes are similar to the estimates in the previous columns. Thus, the estimates in these three columns support the hypothesis that changes in answers about levels reported annually are correlated with past answers about quarterly changes in standards.

The correlation between answers about levels and changes in standards found in the columns 2 and 3 may be driven solely by a correlation between answers about levels and the answers from the most recent year about changes in standards. Indeed, changes in standards reported between October 2013 and July 2014 enter all measures of standards in the first three columns, and they can be responsible for the significant coefficients across all three time periods. Thus, to investigate whether past changes also contribute to this correlation, in column 4, we divide the sum of reported changes over the past two years into the sum over the past year and the sum over the year before and use all four variables as separate controls. The dependent variable is the same one as in column 2, namely the difference between the levels reported in July 2014 and July 2012.

None of the coefficients on the tightening or easing variables is individually significant when the different years are entered separately. However, all of the coefficients have the expected signs, and the four variables are jointly significant at the 6 percent confidence level. Nonetheless, the coefficients on the two variables for reported tightening and easing between the SLOOS of October 2012 and July 2013 are not jointly significant. Thus, we cannot reject the hypothesis that changes in answers about levels in a two-year period are uncorrelated with answers about changes in standards in the first half of this period.

Conclusion
This note used responses to the SLOOS to establish empirical facts about the relationship between annual levels and quarterly changes in lending standards reported by banks. We find that, on average, after controlling for bank and loan fixed effects, the responses to the quarterly survey are loosely correlated with the changes in levels over time. This result suggests that those answers are determined by a common source of information about banks' actual lending standards. However, the weakness or absence of that correlation in simple summary statistics suggests that the relationship is not straightforward and that answers about levels and changes in standards from the SLOOS each contain different relevant information about lending standards. Moreover, the information content of answers about levels and changes in standards could differ for a number of reasons, for instance because respondents may have different thresholds for the extent to which their underlying lending standards must change before they report a change in the survey. Based on these findings, policymakers should remain attentive to answers from banks to both sets of questions.

References
Bassett, William F., Mary Beth Chosak, John C. Driscoll, and Egon Zakrajsek, 2014, "Changes in Bank Lending Standards and the Macroeconomy,"Journal of Monetary Economics, 62, pp. 23-40.

Driscoll, John C., 2010, "Difficulties with Cumulating the Senior Loan Officer Opinion Survey's Changes in Lending Standards Series to Obtain an Index of the Level of Tightness of Standards," memo, Federal Reserve Board.

---

Please cite as:

Bassett, William F., and Marcelo Rezende (2015). "Relation Between Levels and Changes in Lending Standards Reported by Banks in the Senior Loan Officer Opinion Survey on Bank Lending Practices," FEDS Notes. Washington: Board of Governors of the Federal Reserve System, January 16, 2015. https://doi.org/10.17016/2380-7172.1478