«DiscoVeriNg HomelessNess Volume 13, Number 1 • 2011 U.S. Department of Housing and Urban Development | Office of Policy Development and Research ...»
Data The CE Survey consists of two surveys––the Interview Survey and the Diary Survey. The Interview Survey is a quarterly survey of consumer units. This research only uses the Interview Survey. A consumer unit consists of members of a household who are related or share at least two out of three major expenditures. Interviews occur on a rotating quarterly basis; after one household leaves the sample after four interviews, a new household is drawn to replace it, which results in one-fourth of the sample being refreshed every 3 months. Although interviews are conducted on a quarterly basis per household, they are staggered so that households are surveyed every month. In each interview, respondents are asked about expenditures during the past 3 months.
For our research, we used quarterly data from 1984 to the first quarter of 2005; our unit of observation was an owned primary residence. From 1984 to 1998, our sample had 3,100 observations per quarter on average, and, after 1998, we had approximately 5,000 observations per quarter.
The CE Survey provides detailed characteristics for the home and the people living there. The data contain statistics from both the state and metropolitan statistical area (MSA). The home characteristics data include the year the home was built, the type of building (for example, single-family residence), and number of rooms. The respondents also provide the self-assessed home value and details about all outstanding mortgages. Finally, the CE Survey primarily has detailed expenditure data.
As part of collecting the detailed expenditure data, the CE Survey asks questions regarding expenditures related to investments in the home. The CE Survey asks about expenses that occurred during the previous 3 months and creates a separate line item for each improvement project. Some consumer units reported 10 or more separate jobs in a given quarter.1 The CE Survey asks whether each improvement project is considered to be new construction, an addition, an alteration, maintenance and repair, or a replacement.
We follow Reschovsky (1992) in our general classification of expenditures as either maintenance or additions. Maintenance expenditures affect the quality of the existing capital stock of housing.
Additions expenditures add to the capital stock. As Reschovsky noted, this classification is not as clean as one would hope. For example, the replacement of a refrigerator could be classified as maintenance if the new one is of comparable quality. Alternatively, a household may purchase a new state-of-the-art refrigerator that significantly improves the capital stock. We resolve this problem in part by using the consumer unit’s own classification. We classify anything that is coded as new construction or an addition as additions, and we classify anything coded as an alteration, maintenance and repair, or replacement as maintenance.
Summary Statistics Because the CE Survey and AHS have different survey designs, comparisons between the two surveys are not straightforward.2 The biggest difference is that the AHS reports spending on maintenance in a typical year and reports spending on additions over a 2-year period. To get closer to the AHS, we aggregated our quarterly observations and report summary statistics for households that appeared in all four interviews.3 Using the AHS from 1985 to 1993, Gyourko and Tracy (2006) reported that 77 percent of households reported positive maintenance expenditures in a typical year. Exhibit 1 shows that
79.5 percent of households that appear in all four quarters of the CE Interview Survey from 1984 to 2005 reported positive maintenance expenditures.
Using the AHS from 1993 to 1997, Baker and Kaul (2002) found that 16.7 percent of households conducted an expansion project. Our percentage of households spending on additions in a given quarter includes more types of projects than Baker and Kaul included, and we found that a higher percentage of our sample—52.1 percent—spends on additions. This percentage may seem high, but a household can spend $10 on a can of paint that is part of a larger project to add a bedroom The CE Survey asks about the nature of the job and keeps track of the differences over time to help avoid double counting.
See also Rappaport and Cole (2003) for a comparison of the two surveys.
For this comparison, we lose 40.4 percent of consumer units (but only 18.7 percent of quarterly observations) by restricting the sample to households that appeared in all four interviews.
Tracy (2006), exhibit 1, which used data from 1985–1993. The unconditional mean values come from Davidoff (2006), exhibit 1, which used data from 1985–2001. The additions value is divided by two because Davidoff reports a value over 2 years, as it is reported in the AHS.
Notes: The Consumer Expenditure (CE) Survey results are based on authors’ calculations using data from the first quarter of 1984 to the first quarter of 2005. The yearly observations column for the CE Survey restricts the sample to those households that appeared in all four interviews. The rest of the article uses the quarterly data.
and have it count as an addition if the household identifies it as such. If a household is slowly performing a project, the costs can be spread out over time.4 Using the AHS from 1985 to 2001, Davidoff (2006) found that households spent $553 and $1,793 per year on average on routine maintenance and additions, respectively. Exhibit 1 shows that our sample spent $1,110 and $1,147 per year on average on maintenance and additions, respectively.5 Summing over the two categories, Davidoff found that households spend $2,346 per year on maintenance and additions (in 2003 dollars), and we found that households spend $2,257 per year on maintenance and additions (in 2004 dollars).
These cross-sectional values in the AHS and CE Survey mask considerable time-series variation.
Returning to our sample of quarterly observations, the percent spending on maintenance peaks in 1984 at 52 percent per quarter and declines to 42 percent by 2004 (exhibit 2).6 The percent spending on additions shows a different pattern, starting at 23 percent in 1984 and declining to
17.5 percent in 2000 before increasing to more than 21 percent in 2004 (exhibit 3).
Although fewer households spent on maintenance and additions over time, mean quarterly expenditures were constant or increasing. Real mean maintenance expenditures increased from $230 to $320 between 1984 and 2004 (exhibit 4), suggesting that households were doing fewer but larger projects. There is considerable variation in mean additions expenditures, with a noticeable We censored the data and only counted additions if the household spent at least $1,000, and the trend in additions did not change.
The expenditure data are in real 2004 dollars, using the Consumer Price Index research series.
The data are seasonal. We smooth the figures using a simple four-quarter moving average.
decrease in mean expenditures after 1988, and then a gradual increase after 1995 and a dramatic increase after 2000. The first 4 years of the 2000s is particularly interesting for additions, because the proportion spending on additions increased dramatically (from 18 to 21 percent per quarter) and mean spending increased $300 to $450 per quarter. Households were more likely to spend money on an addition and, conditional on spending, to spend more as well.
Exhibit 2 Fraction With Maintenance Expenditures, First Quarter of 1984 to First Quarter of 2005 0.55 0.50 0.45
Unconditional Mean Quarterly Maintenance and Additions Expenditures, First Quarter of 1984 to First Quarter of 2005 Expenditures ($)
Empirical Methodology To explore the stylized facts of the literature with the CE Survey, ideally we would match the empirical methods used in existing research. In the broadest sense, following the existing literature means we would estimate reduced form models, studying the determinants of the decision to spend and the determinants of the level of spending. Given that we are estimating reduced form models, the first methodological challenge posed is that observations with zero expenditures are numerous. The specific methods employed differ across authors, including ordinary least square (OLS), independent two-stage estimation, and two-stage Heckman selection correction.
In this study, we used an independent two-stage estimation strategy. We first used a probit to model whether the household spent during the quarter, and then we model log expenditures as a linear function via OLS. Estimating the two stages separately required the assumption that the errors from the probit model and those from the OLS model are independent. A large portion of both maintenance and additions expenditures in our data represent large and lasting projects with a significant durable-goods character. Such long-term investments are not likely to be considered anew by the household every period. Instead, many of the expenses are likely to be triggered by some exogenous process (for example, the roof leaks), which forces the household to make a maintenance decision. These external factors are likely to be uncorrelated with the cost of the maintenance.
One important way our study differs methodologically from the existing research is that we estimated the extent of maintenance and additions expenditures separately but estimated the decisions simultaneously. Most research pools maintenance and additions expenditures together, while, in this study, we allowed the independent variables to affect the two expenditures differently. We believe that the decision to maintain the capital stock can be different from the decision to add to
the capital stock. For example, a large addition project might be more sensitive to the interest rate than a routine maintenance expenditure.
We estimated a bivariate probit model for maintenance and additions; then we estimated an independent, seemingly unrelated regression for log expenditures. The dependent variables in our bivariate probit model are whether the household spent a positive amount on maintenance and whether the household spent a positive amount on additions. The dependent variables in our reduced form demand model are log of maintenance expenditures and log of additions expenditures.
Independent Variables The independent variables in each regression are identical.7 For the demographic characteristics, we included log before-tax income, family size, and age of the respondent. For characteristics of the home, we included dummy variables for home age, rooms, urban or rural status, and MSA status. We augmented the CE Survey data with additional macrolevel variables. The regressions included month dummy variables to capture the seasonality of expenditures and three aggregate variables to capture possible macroeconomic effects. We included the 30-year fixed mortgage interest rate for each quarter from Freddie Mac to capture differences in borrowing costs.8 To capture changes in housing markets, we used the Office of Federal Housing Enterprise Oversight (OFHEO) (now the Federal Housing Finance Agency), repeat sales house price index at the state level. We also included the state employment rate to capture the potential effect of labor market conditions.
Empirical Results Exhibits 5 and 6 present the results. We first discuss our results relative to the stylized facts.
Interest Rates The first stylized fact is that interest rates play a role in the maintenance and addition decisions.
Many papers model the household’s decision to increase the size of its home as a choice between making additions to the current home or moving to a different home (for example, Potepan, 1989). In these models, a household finds itself with extra demand for housing services as the result of some exogenous shock and meets that demand either by adding to its current house or by moving. Because most houses are financed, interest rates play a critical role in the relative costs and benefits of each option.
Potepan (1989) creates a model of household choice in which a household stays put when prevailing mortgage rates are high relative to the household’s mortgage rate. Potepan describes it as the mortgage lock-in effect, indicating that a household with an interest rate lower than the current market rate is locked into its current home because it would be too expensive to move to a new home at the prevailing interest rates. Thus, a higher interest rate may lead to a higher probability of an owner staying in the home and spending money on maintenance and additions, although the maintenance and additions may also need to be financed with a loan.
See the appendix table for summary statistics for the independent variables.
We found that an increase in the 30-year OFHEO mortgage interest rate increases the probability of maintenance and additions expenditures (exhibit 5). Thus, we found evidence for the mortgage lock-in effect. Similarly, an increase in interest rates increases the level of spending on maintenance and additions projects (exhibit 6).
Income The effect income has on maintenance and additions expenditures is complicated by the decision to move. Because housing is a normal good, one might assume that maintenance and additions might be normal goods as well. A large amount of the research that includes income as a variable found that an increase in income leads to an increase in maintenance and additions expenditures.
This simple explanation ignores the more complicated relationship that occurs when a household
moves to alter its housing. Montgomery (1992) found that an increase in income increases the likelihood of moving and the likelihood of improving its current home relative to doing nothing.
We do not explicitly include the decision to move from or stay in a home as a variable, so we have to be careful in interpreting our results in light of this omission. As with the existing literature, the results of this study suggest that an increase in income increases the likelihood and level of maintenance and additions expenditures. The effects are larger for maintenance, especially for the dollar value of maintenance expenditures.