Oswaldo Molina July 1, 2008

1. Impact of property rights on poor households’ investment decisions: a treatment evaluation of a titling programme in Peru Oswaldo Molina July 1, 2008

2. Contents • Motivation • Data • Methodology of impact evaluation • Defining control groups and potential bias problems • Empirical estimation • Empirical results • Baseline results • Robustness to functional form • Dynamic response • Final remarks

3. Motivation Why this topic can be interesting? (1) • Protection of property rights has long been emphasized as an essential precondition development (North and Thomas, 1973; Demsetz, 1967; Johnson et al., 2002). • Fragile property rights not only tend to reduce total investment but also have significant effects on its composition. Tenure insecurity hinders long-term investments(Dercon et al 2005). • Nowadays, millions of people in urban areas of developing countries occupy dwellings (31.6% of the global urban population). A large proportion without a title. • This topic has become primordial on the policymakers’ agenda (Baharoglu, 2002; Field, 2003).

4. Motivation Why this topic can be interesting? (2) • Many governments have started land-titling programmes in urban areas (such as Colombia, Mexico, Peru, Angola, Senegal, South Africa, India). • Even though a considerable empirical literature explores the effects of property rights on investment, it has been principally focused on rural areas. • The Peruvian experience is one of the largest government titling programmes targeted to urban areas (more than 1.5 million property titles were recorded by governmental agency Cofopri) (Cofopri, 2006).

5. Motivation Objective of this paper • To evaluate the impact of the Peruvian large-scale titling programme on housing investment. • The Peruvian experience was previously analyzed by Field (2005): impact of titling is limited only to short-run investment. • Some of her findings contrast those of this paper: we find not only a positive relationship with short-run investments, but also with long-run ones. • This analysis considers the methodology suggested by Field (2005) as a starting point using. We extend the analysis utilizing different econometric techniques, employing different control groups, dealing with endogeneity problems and including a richer set of control variables.

6. Data Data (1) • Cross-section data set, collected in June 2003 from five different regions (includes panel data information of eight categories of housing investment). • It includes information of tenure status from 2331 properties (836 having a Cofopri’s title). 51% are communities that were effectively reached by Cofopri. • Ex-post cross-section data can be used to evaluate programmes if (Field and Kremer, 2005) • it incorporates retrospective questions about the intervention, • data cover enough period to estimate the total benefits. Fortunately our survey satisfies both requirements.

7. Data Data (2) • Defining the investment variable. • Before-programme: sum of the number of investments undertaken in the two years priors to the programme. • After-programme: sum of number of investments completed in 2001 and 2002. • It is also feasible to distinguish between short-run and long-run housing investment. • Investment variable has some specific characteristics

8. Methodology of impact evaluation Defining control groups and potential bias (1) • Two different control groups are used to provide more robustness to our results. • First control group: households in communities that were reached by the programme and that did not obtain a title, because they did not fulfil all the requirements. The selection is at the household level. • Second control group: households that, according to requirements, were eligible to get a title, but did not get one because they lived in areas that were not treated yet by Cofopri (potential future beneficiaries). The selection is at the area level.

9. Methodology of impact evaluation Defining control groups and potential bias (2) Potential biases: • First control group (household level): selection bias. • The analysis incorporated as controls the requirements to obtain a title (residency time and non-possession of other proper title). • Second control group (area level): timing in which Cofopri reached each community is related to any unobservable variable that, at the same time, is correlated with investment. • Not contaminated by the potential selection bias of the first control group • Programme seems to focus first on the easier to title lots (average cost of titling increased over time; Morris, 2004). Timing in the implementation was not exogenous. • Analysis also includes the variables that were considered in the selection of the cities (distance from commercial centres, city size and concentration of informality)

10. Methodology of impact evaluation Empirical estimation • The expression for the investment level: • After taking first differences becomes: • This strategy allows us to remove any bias produced by time-invariant unobserved heterogeneity as it cancels out upon subtraction.

11. Empirical results Baseline results (OLS difference-in-difference models)

12. Empirical results Baseline results (2) • Results using OLS models, similar to the methodology employed by Field (2005). • Large impact of Cofopri’s title on total housing investment. Being treated implies that the expected number of investments increases by 0.20-0.30 (rises by 60% on average). Results are similar to those of Field (2005), whose reported treatment effects at 68%. • Average treatment effect on long-run investment is about 0.08 and highly statistically significant (given the low baseline, an increase by 0.08 implies an increase by more than 200%).

13. Empirical results Baseline results (3) • These results differ substantially from those obtained by Field (2005). There are (at least) two reasons for this: • Our regressions include a richer set of control variables. • Perhaps more importantly, our data span a longer period after titling than Field’s data. • Impact of the programme by level of income. To do so, we estimate regressions for each quartile of income. Results indicate that as the level of income increases, the significance and the coefficient associated with the impact of titling also rises, especially in long-run investment. • These results suggest that other barriers exist, besides risk, which limit investment for the poorer households, and can be then attributed to persistent market failures. • These programmes need to be complemented with other policy measures.

14. Empirical results Robustness to functional form (count data models) • Results tend to be lower than those obtained in the OLS models. Impact on total investment is between 0.17 and 0.26.

15. Empirical results Robustness to functional form (dif-in-dif propensity score matching models) • Two different propensity score, according to the control group (household’s prob. of being selected and community’s prob.) • Low bias if we incorporate in the participation regression the variables which explain selection (Heckman et al, 1997) • Impact on long-run investment: an increase in the number of housing sizeable additions by 170-200%.

16. Empirical results Dynamic response • Although we do know that title impact positively on investment, we do not recognize if this impact tends to be immediate or if it takes time to be relevant. • We construct the temporary investment behaviour of each region. Considering as time zero the two years prior the treatment, we generate a binary variable of any investment in two-year periods and compare each of them with the pre-programme baseline. • In the case of total and short-run investments, the impact of title on housing renovations is significant even in the following two years after the programme. • On the contrary, title enhances the probability that a household makes a long-run investment, but only four years after of being treated. Households appear not to react promptly to the incentive provided by the title.

17. Empirical results Dynamic response • A considerable horizon of time is required in order to measure the complete impact of a titling programme.

18. Empirical results Final remarks • Title presents a highly significant and larger effect on long-run investment. The results from the dif-in-dif propensity score matching indicate that the estimated average treatment effect implies an increase in the number of housing sizeable additions by 170-200%. • Impact of the programme is different depending on the level of income: as the household’s income increases, the significance and the coefficient associated with the effect of titling also rises, particularly in long-run investment. • While the effects on housing renovations can be significant even in the following two years after the implementation, its impact on long-run investment requires more than four years. This result has serious implications for the evaluation of programmes: a considerable time-horizon is needed to measure its total impact. • Collecting new panel data sets can allow further research to produce more accurate estimations. • Other ideas: anticipation bias, differential investment behaviour associated to risk aversion and split total impact in its components.

19. Impact of property rights on poor households’ investment decisions: a treatment evaluation of a titling programme in Peru Thanks.