Paul Gompers EMCF 2009 March 5, 2009

Transcription

1 Paul Gompers EMCF 2009 March 5, 2009

2 Examine two papers that use interesting cross sectional variation to identify their tests. Find a discontinuity in the data. In how much you have to fund your pension plan. In how easy

3 Josh Rauh JF 2006

4 Sponsors of defined benefit DB pension plans need to make contributions to their pension pools that are legally specified, i.e., formulaic. DB plans pledge future payments to workers Based on retirement cohorts, promised benefits, assumed returns, and existing assets, pension plans are either under-, fully-, or overfunded. Funding status has non-linear affects on amount of corporate contributions to the pension fund. This funding amount varies from year to year for a variety of reasons.

5 Funding requirement can vary because: Fund returns Changes in the discount rate applied to future benefits. Voluntary funding decisions. Changes in the future benefit structures.

6 Does the required level of contributions to the pension fund (Mandatory contributions MC) affect the capital expenditure and R&D of public companies? Changes in the contribution may be correlated with firm prospects, its cash flow, and its investment opportunities. i But There are non-linearities in the contribution requirements for certain levels of underfunded status. Also for overfunded pensions, lose tax benefits of contributions.

7 Estimate the effect using these non- linearities. Get identification of whether external finance is costly. Instrument for internal cash.

8 Figure 1 Funding status of all public traded firms on Compustat. Big cross sectional and time series variation. Data from SEC filings. Unfortunately, SEC filings don t give enough information to run test. Get plan information from IRS form 5500 filings. Hence, sample runs from 1990 to 1998.

9 Underfunded plans must contribute amount equal to new benefits accrued plus a fraciton of the funding shortfall. Contribute the larger of: Minimum funding contribution (MFC) Deficit reduction contribution (DRC) Prior to 1994 = min{0.30, [ *(funding status- 0.35)] After1994 = min{0.30, [ *(funding status-0.60)] Figure 2. Overfunded plans are not required to contribute. Can make voluntary contributions, but big Can make voluntary contributions, but big contributions lose tax advantage.

10 Typical regression of investment on Q and cash flow. Include Z it, mandatory contribution (MC) Include firm and year FE

11 Still endogeneity concerns, but will utilize Still endogeneity concerns, but will utilize the kink in MC to identify exogenous variation in internal cash.

12 Unbalanced panel using the IRS 5500 filings for all firms that report DB pension assets. Typically older manufacturing firms. 8,030 firm-year observations for 1,522 firms. Table I Summary stats. Compare actual to required contributions. Figure 3. Use a kernel estimation of the density function. Definition of The Epanechnikov Kernel: The Epanechnikov kernel is this function: (3/4)(1-u2) ) for - 1<u<1 and zero for u outside that range. Here u=(xxi)/h, where h is the window width and xi are the values of the independent variable in the data, and x is the value of the scalar independent variable for which one seeks an estimate.

13 Graph mandatory contributions/book assets Figure 4

14 Look at relationship between funding status and capital expenditure and funding status and pension contributions, both scaled by assets. Do non-parametric univariate analysis using the Epanechnikov kernel. Figure 5. Looks as if there are similar inverted patterns in capex and pension contributions.

15 Regression with a variety of specifications. Baseline without MC. Breakout contributions into mandatory and total contributions. Look at funding status of pension fund. Cluster standard errors by firm to correct for within firm serial correlation of error terms. Alternatively, cluster by year (ala Fama-McBeth) or use an arbitrary AR(1) correction in the panel. Table II. Only mandatory contributions matter. Include funding status as an independent variable. Hence, it is the shape of the contribution curve that identifies ifi the effect, not funding status.

16 Can instrument for pension contributions by using MC as an instrument for either total pension contributions or total firm cash flow. Identifying that component of pension contribution or cash flow that is correlated with MC. Include funding status as an independent variable. Hence, it is the shape of the contribution ti curve that identifies the effect, not funding status. Table III.

17 Examine a variety of other uses of corporate cash. R&D, Acquisitions, dividends, repurchases, and changes in debt. Table IV. No affect on R&D. Perhaps R&D has big startup and stopping costs. Reduction in acquisitions. In Tobit regressions, reduction in dividends and repurchases. Increase in debt. Reduction in trade credit. Increase in working capital.

18 Divide MC into the unexpected and expected components. Need to calculate the expected pension assets and liabilities. i Utilize information on pension liabilities, expected return (based upon share of stock and bonds in portfolio). Table V. Both expected and unexpected MCs matter.

19 Prior literature has utilized a variety of measures of funding constraint proxies. These have been shown to be related to cash flow-investment sensitivities in a variety of papers. Utilize: Age Credit rating Dividend ratio Cash balances Those firms with Capex > cash flow. Table VI. In general, variables that are proxies for financing constraints t are associated with greater sensitivity of investment to cash flow.

20 Is there a cost to the firm of MC? Level of investment around large MCs (>0.1% of book value of assets). Figure 6. Do other firms take up the slack? Estimate total industry pension requirements. Table VII: Look at firms based upon cash and capex to CF. Unconstrained firms seem to increase investment.

21 Kink in MC based on funding status is a nice natural variation that allows for examination of Cash flow-investment puzzle. Financing constraints do seem to matter for these types of firms. Remember, these are DB companies which h may have very different types of investment behavior.

22 Benjamin Keys, Tanmoy Mukherjee, Amit Seru, and Vikrant Vig Working paper 2008

23 Incentive problems potentially exist when banks/lenders originate loans and then sell them in a securitized pool. Perhaps less incentive to fully gather information if you are selling the loan off. Papers have looked at this issue, but endogeneity issues plague most of them. Exploit an industry rule of thumb to identify any effects on monitoring/information gathering.

24 Prominent rule of thumb is to not lend to people with FICO scores less than 620. FICO scores FICO is the acronym for Fair Isaac Corporation, a publicly-traded corporation (under the symbol "FIC") that created the best-known and most widely used credit score model in the United States. Calculated statistically, with information from a consumer's credit files. Primarily used in credit decisions made by banks and other providers of secured and unsecured credit. Intended to show the likelihood that a borrower will default on a loan Range is 400 to 900.

25 This rule of thumb makes securitizing loans of borrowers with FICO scores less than 620 more difficult, i.e., they are less liquid. Guidelines by Freddie Mac Cautious Review Category Research design: Look at borrowers just above and just below the 620 break. Should have similar default probabilities. Look at quality of the loans, terms, etc

26 Borrowers on either side of 620 should look similar. Only small differences in characteristics. Screening is costly for lenders.

27 Look at loans where collection of soft information may be important. Low documentation loans. More likely that soft information would be important to estimate default probabilities. Full documentation loans.

28 60% of loans trade as mortgage-backed g securities. Most are agency-pass through pools. Freddie Mac (Federal Home Loan Mortgage Corporation), Fannie Mae (Federal National Mortgage Association), Ginnie Mae (Government National Mortgage Association). Agency eligibility is based upon loan size, credit score, and underwriting standards. Implicit government guarantee. Non-agency loans subprime More expensive. Price of loan depends upon credit score, debt to income, and documentation level. No guarantee of loan.

29 LoanPerformance. Detailed data on non-agency securities markets. 8,000 home equity and nonprime loan pools millionloans. $1.6 trillion outstanding. 90% of all securitized subprime loans.

30 Hard information: FICO Score Loan terms (LTV, interest rate) Soft information: Measure of future income stability of borrower, years of information provided by borrower, joint income status). When securitized, only hard information, y provided.

31 Y is number of loans of score I T = indicator if FICO>620 and 0 if FICO<620. T*f(FICO) is a flexible seventh-order polynomial. Fit smooth curve. Data recentered so that FICO = 620 is 0. At cutoff, polynomial is evaluated at 0. β is measure of discontinuity for FICO>620.

32 Table I Focus only on low documentation loans. Those where soft information will be important. Figure 2 Increase in number of loans above 620. Yearly. Table 2 β coefficient by year. Do permutation test. Look for discontinuity at other places. Allow the 0 to be at different FICO scores. Do not find any other discontinuities.

33 Look at LTV and interest rates. Figures 3 and 4.

34 Look at whether or not performance of loan differs around 620. Look at default rates. Dollar weighted. Default within months of origination. Collapse data into 1 point FICO bins. Figures 6A-F. Table III. Figure 7 Delinquencies by age of loan.

35 Look at loans on either side of 620. Group for loans 615 to 619 and T=1 if FICO is between 620 and 624 and 0 for FICO between 615 and 619 between 615 and 619. Control for type of loan Adjustable or Fixed rate. Age. Logit is Panel C of Table III.

36 Look at Georgia and New Jersey. Both passed Fair Lending Laws. Strong restrictions on predatory lending. Made securitizing loans very difficult when law was in effect. Both laws later repealed. Run same regression and include interaction for when Fair Lending law was in effect and not. Table IV. Big increase in loans above 620 when law not in effect in those states. No effect when Fair Lending law in place.

37 Do borrowers manipulate their FICO score to be just above 620 Fair Isaac says that it takes time and is hard to do. Look at six months immediately after repeal of Fair Lending laws. Table IV Panel B. Change in delinquencies happens immediately Change in delinquencies happens immediately after law for loans with FICO > 620.

38 Does the effect exist in full documentation loans. Greater information about borrower s ability to repay. Fair Isaac advises lenders that below FICO of 600, very troubled borrower. Figure 11 Substantial increase in full documentation loans with FICO>600. Look at default rates above and below FICO 600 for full documentation loans. Figures 12 and 13. Table VI.

39 Seems as if loan default differential in low documentation loans is due to soft information For full documentation loans, i.e., with more hard information, no difference at loan inflection point.

40 Interesting use of kink in the behavior of lenders. Find that this leads to an effect on loan repayment. Nice set of robustness tests.