This chapter is divided into four sections. The first section discusses the general characteristics of the 1979 HUD-1 survey. The second section presents the results of the current (1979) survey. A comparative analysis of the 1979 and 1975 HUD-1 data sets constitutes the third section. Concluding remarks on price impact close the chapter.

GENERAL CHARACTERISTICS OF THE 1979 HUD-1 SURVEY

This section presents a brief overview of the 1979 survey of HUD-1s conducted by PMM&Co. A more detailed discussion is included in the introductory chapter of Volume III of this report.

Survey Overview

The 1979 HUD-1 survey collected HUD-1 data from a stratified random sample of U.S. institutional lenders. Lenders were stratified along two dimensions: lender type and lender size. There are four types of institutional lenders:

Lenders within each institutional type were divided into large (stratum I) and small (stratum II) lenders. Each large lender accounted for 10 percent or more of the loan origination activity for its geographic area.

Randomly selected lenders from each type and size category in each of the 50 states and eight standard metropolitan statistical areas were asked to submit HUD-1s. The SMSAs are:

In order to derive reasonably reliable estimates of settlement costs, a threshold of at least 250 observations per state and SMSA was specified. Oversampling by a factor of four allowed the study team to attain at least this number of observations for nearly all geographic groups. These thresholds; together with the expected level of loan origination activity for an area determined the number of "months' worth" of HUD-1s requested from lenders. The large majority of HUD-1s were records of closings between October 1978 and January 1979.

Potential Inaccuracies And Biases

It is important to note that although the HUD-ls are drawn from a representative sample of lender types, lender sizes and geographic areas, there is the potential that they contain certain biases and inaccuracies. In addition to the usual clerical and coding errors commonly encountered in survey sampling and normally of minimal importance, at least two other potential sources of error may tend to produce distorted estimates. The first involves the notion of construct validity; the second is self-selection bias.

Construct Validity

"Construct validity" refers to the extent to which the sampling instrument (in this case the HUD-1s) accurately reflects the variables the survey is attempting to measure (closing costs). It became clear through conversations with lenders and HUD personnel alike that there are considerable definitional and interpretive problems with the line items on the HUD-1s. For example, it is not altogether clear where particular costs are to be entered on the form, as local nomenclature may differ somewhat from that appearing on the form. Furthermore, total closing costs are often not fully disaggregated into their various component costs. Frequently, one type of cost is subsumed under and included within some other cost and it is difficult to separate such costs into their proper parts post hoc.

As settlement practices vary from area to area, so do filing practices. For example, banks in some areas do not always complete the forms, and there is no incentive to ensure that other providers have entered accurate data. Moreover, the study team learned that the interpretation of the line items varies not only from area to area but also within a given institution. is, filers may file differently even within the same institution.

Self Selection Bias

That

In addition to the problems raised under construct validity, the representativeness of the sample of all transactions completed in the areas studied and within the timeframe specified may also be questioned due to self-selection biases. Although its influence cannot be determined accurately and precisely, non-respondent bias, that is, refusal by a selected lender to participate in the study, may obfuscate sample estimates. Thus, it is possible that the responses obtained may represent a different distribution of lender types than prevails generally in the housing markets studied.

Secondly, responding lenders may have engaged in some degree of selfselection of the forms they submitted. Although selected lenders were asked to send forms for all transactions closed during the test period(s), and most probably did so, there was no way to ensure that they did so without some kind of on-site check of the lender's files. For example, a particular lender might have eliminated some forms which showed that he charged unusually high settlement fees on the grounds that these forms would reflect adversely on him. 1

Nonresponse and self-selection biases are therefore potential impediments to obtaining reliable estimates and they may be present in the 1979 HUD-1 data set. We suggest, therefore, that the estimates presented here be compared and used with care.

RESULTS OF THE 1979 SURVEY OF HUD-ls

This section presents estimates of average settlement costs by geographic area, price range, and cost class. Sample estimates are derived for three geographic groups:

Exhibit XV-2 presents national averages for the following classes of settlement costs:

Items payable in connection with the loan (Series 800);

Items required by the lender to be paid in advance (Series 900);

1Because a limit of 250 forms from a given institution was specified, large lenders may have engaged in even more exaggerated forms of self-selection. 2Contract sales price is not a settlement cost. It is included here for normalizing settlement changes across geographic areas.

Government recording and transfer charges (Series 1200);

The estimates include charges to both the home buyer and home seller. Estimates are computed using only nonzero HUD-1 entries. That is, the number of observations used in computing sample means and standard deviations is the number of nonzero costs appearing on the HUD-1 for the given class of cost. For example, suppose only 1,000 of the some 18,000 HUD-ls contain a broker's commission; then the sample mean and standard deviation for broker's commission would be computed using an N (the number of observations) of 1,000, not 18,000. This implies, of course, that summing the averages of the component settlement costs will overstate average total settlement costs. That is, the sum of the average component costs will not equal the average total settlement costs. alternative, however, of including zero costs in the averages is less attractive as it results in seriously understated average component settlement charges. Estimates are derived after removing all "outliers" beyond four standard deviations from the mean.

The

The national estimates are computed using the full set of HUD-1 data, which includes all the state and SMSA data. No attempt was made to precisely weigh the observations of a given state based on the state's share of the nation's aggregate level of origination activity, as these data were not readily available. Rather, the estimates reflect essentially equal weightings of each state supplemented by the data from the SMSAS. The estimates presented, threfore, are not true national estimates but an approximation of those estimates.

A review of Exhibit XV-2 yields the following key findings:

Average contract sales price for the national data set is
$54,864.

1Broker fees are not included in total settlement charges for two reasons: (1) wide variation across geographic units in the incidences with which these fees are recorded on the HUD-1 may distort the resultant estimates; (2) broker fees are very large relative to other settlement charges and they would therefore swamp individual settlement costs if included in the totals.

2Taxes include all government recording and transfer charges and all state

and local taxes.