Presentation is loading. Please wait.

Presentation is loading. Please wait.

Discussion of “Foreclosures In Ohio: Does Lender Type Matter?” Robert B. Avery January 2, 2009.

Similar presentations


Presentation on theme: "Discussion of “Foreclosures In Ohio: Does Lender Type Matter?” Robert B. Avery January 2, 2009."— Presentation transcript:

1 Discussion of “Foreclosures In Ohio: Does Lender Type Matter?” Robert B. Avery January 2, 2009

2 2 Question: Is variation in mortgage delinquency and foreclosure rate across 88 Ohio counties due to differences in mix of lender types? What is best type of analysis?  Reduced Form  Structural Model  Trigger Events  Put option on the Collateral  View failure to default as exercise of option to buy property

3 3 Assumptions under different structural models Trigger event  Best viewed as a model of delinquency  Life events—job loss, divorce, health problems— cause household to be unable to make payments Contingent Claims models  Better model of foreclosure or nonowner-occupied owner  Presumes owner action is a choice not a necessity

4 4 Aggregating to county-level data Trigger event  Markov process  Not easy to model loan seasoning or state dependence—can be observationally equivalent. Contingent Claims models  Critically dependent on assumptions about distribution of house prices and original LTVs.  Foreclosures in particular likely to follow very non- linear aggregation patterns—no foreclosures in some years; lots in others.

5 5 How well does author’s data suit structural model aggregation assumptions—control variables? Average house price data from property tax files—uses levels of house prices not changes. Mortgage value proxied by median monthly owner costs from 2000 census—no time variation. Trigger event data—also proxied by 2000 census variables Time varying data include divorce rate, unemployment rate, income growth (from REIS??), call data average on arms and restructured mortgages

6 6 How well does author’s data suit structural model aggregation assumptions—critical variables? Foreclosure filings (annual) from county records divided by owner-occupied units. Delinquency from call data of only regulated lenders allocated by distribution of HMDA originations Non regulated lender share from single-lagged total HMDA loans (or dollars) further separated by home purchase/refi. Not clear if it includes non owner-occupied or 2 nd liens.

7 7 Issues Too many proxies and non-time varying variables to support structural view—also not clear which structural model at play. Lack of home equity data is a critical omission for contingent claim model and foreclosure equation (Massachusetts data used by many authors). Timing of tax data on house pricing of concern as well as lack of change data. Single lag in data hard to generate from any theory. Estimate from June 2007 credit bureau data the following chart for Ohio:

8 8 Issues (continued)

9 9 Hard to interpret the meaning of the lender variables when many endogenous variables included on the RHS—particularly delinquency For example, if nonbank lenders make weaker loans more likely to go delinquent all else equal, entire effect could be subsumed in the delinquency variable showing no effect on foreclosures when there really was one. Results suggest that lender type effects foreclosure but not delinquency. Hard to see how this can happen. Argues for reduced form.

10 10 Issues (continued) Variable for delinquency has many problems. Allocation method makes sense only for very small banks (and should take account of loans sold). Crucially it ignores loans originated by non- banks. Could actually have opposite sign as expected (consistent with real results). This might explain result of no impact on delinquency Credit bureau (Trend Data) much better alternative.

11 11 Issues (continued) Problem with the sample HMDA centered on MSA reporting—lender must have office in an MSA. Underreports smaller, local lenders in rural areas. For example, in 2004, 37 percent of banks and thrifts with an office in Ohio did not report HMDA. Further 48 of 88 counties were rural. At the very least would want to see results only for MSAs.


Download ppt "Discussion of “Foreclosures In Ohio: Does Lender Type Matter?” Robert B. Avery January 2, 2009."

Similar presentations


Ads by Google