A large panel data of bilateral FDi outflows to 119 developing economies from 27 developed econo-mies over the period 1985−2012 was used to examine the effect of biTs on FDi to developing economies.

The modified gravity equation was estimated based

on two estimation methods: ordinary least squares (olS) and Poisson pseudo-maximum likelihood (PPMl). All time-variant explanatory variables were lagged by one period to reduce endogeneity problems.

**Ordinary least squares (OLS)**

• Given the multiplicative form of the gravity
equation, the usual method is to take the natural
logarithms of the explained and explanatory
variables (excluding dummies) and apply ordi-nary least squares to the resulting log-linear
equation.^{2}

• To control for omitted variable bias, home and
host fixed effects were included through dummy
variables which control for all time-invariant
home or host country characteristics.^{3} Also
included were time fixed effects to account for
any shocks that affect all countries.

• Columns 1 to 5 of the table 6.A.1 present the results of the estimations obtained by olS, along with robust standard errors and three types of fixed effects (year, host country and home country). overall, this specification explains about 50 per cent of the variation of bilateral FDi outflows. Results show that except for openness and common border, coefficients are all statistically significant. in particular,

“geographical distance” has a strong effect: its negative sign indicates either that FDi is related to bilateral trade or high operating costs due to geographical distance, and cultural and insti-tutional differences. The coefficient of “labour skill” in host countries has a positive sign, suggesting a more important role of domestic markets. All other variables have the expected sign. in this specification biTs coefficients are significant and positive. However, the propor-tion of FDi that can be attributed to biTs is very low, as reflected in negligible change in R-squared when including a biT variable.

**Poisson pseudo-maximum likelihood ****(PPML)**

• Santos Silva and Tenreyro (2006) showed
that due to Jensen’s inequality^{4}
the use of log-linearized gravity models by olS can generate
biased^{5}
estimations and produce misleading con-clusions. They suggested that the coefficients in
the gravity equation should be estimated in its

multiplicative form, and proposed using the Poisson pseudo-maximum-likelihood (PPMl) estimation method. PPMl is consistent in the presence of heteroskedasticity, and provides a way to deal with zero values (unlike logarithm specifications).

• Columns 6 to 10 show results obtained by PPMl, along with robust standard errors and three fixed effects. The coefficient of skill differ-ence is statistically significant, and its positive sign provides support for FDi that is motivated by lower wage costs in the host country. Market size, labour skill, openness and RTA are all statistically significant and have the expected sign, whereas coefficients of biT variables are not significant. The coefficients of the four time-invariant variables – geographic distance, common border, common language and colony – are all statistically significant.

• Ruiz and Vilarrubia (2007) argue that because
cultural and historical factors are difficult to
measure, gravity models should be estimated
by using time and country-pair^{6} fixed effects.

Columns 11 to 15 show the results of the esti-mations by PPMl, with year and country-pair treated as fixed effects. except for biT varia-bles, all time-variant coefficients are statistically significant. Sizes of coefficients are, in general, higher than those obtained by PPMl with year, home and host country fixed effects.

• When comparing results with those obtained using the olS specification, olS estimates tend to be much larger than those estimated by PPMl. This shows that results are quite sensi-tive to the specification. For this reason, the results of previous studies using olS should be interpreted with caution.

• To check for robustness, the gravity equations were also estimated by including alternatives definitions of variables such as openness (i.e.

total trade over GDP), skill difference (i.e. abso-lute value, positive and negative values), and
biT (i.e. number of years since ratification of a
biT). Moreover, various transformations of the
FDi variables were tried.^{7}
in all these specifica-tions the PPMl estimates of the coefficients of
biT remained statistically insignificant.

**Table 6.A.1****REGRESSION RESULTS, 1985–2012** *(Bilateral FDI, millions of dollars)* *OLS: ln FDI**Poisson-Pseudo-Maximum Likelihood (PPML): FDI* *Explanatory variable**1**2**3**4**5**6**7**8**9**10**11**12**13**14**15* ln GDP - host0.93***0.96***0.90***0.88***0.94***1.16***1.19***1.16***1.15***1.16***1.22***1.23***1.22***1.21***1.22*** ln GDP - home2.40**2.33**2.42**2.41**2.40**0.97**0.93*0.97**0.98**0.97**1.40**1.40**1.41**1.40**1.40** ln labour skills - host3.94***3.89***3.90***3.87***3.95***1.58***1.48***1.60***1.57***1.58***1.93***1.91***1.88***1.88***1.93*** ln skill difference2.57***2.58***2.48***2.39***2.59***0.95**0.96**0.98***0.90**0.94**1.44***1.46***1.36***1.32***1.40*** ln openness-0.10-0.11-0.13-0.14-0.100.36**0.40**0.36***0.35***0.36***0.45***0.47***0.44***0.43***0.43*** Regional trade agreement0.35**0.42***0.20** BIT - signature0.25**-0.040.09 BIT - entry0.44***0.060.14 BIT - years since signature0.000.000.01 * Time-invariant variables* ln distance-2.61***-2.57***-2.59***-2.58***-2.62***-0.73***-0.68***-0.73***-0.73***-0.73*** Common border-0.06-0.10-0.000.00-0.060.53**0.47**0.52**0.54**0.52** Common official language2.54***2.54***2.56***2.56***2.53***0.92***0.88***0.92***0.94***0.93*** Colony1.51***1.51***1.48***1.46***1.51***0.26**0.31***0.27***0.24**0.25** Number of observations12 57312 57312 57312 57312 57312 57312 57312 57312 57312 57312 57312 57312 57312 57312 573 R-squared

*0.5000.5000.5000.5000.5000.7340.7370.7340.7340.7340.8250.8260.8250.8250.825*

**a**

**Fixed ef***YearYesYesYesYesYesYesYesYesYesYesYesYesYesYesYes Host countryYesYesYesYesYesYesYesYesYesYes Home countryYesYesYesYesYesYesYesYesYesYes Country pairYesYesYesYesYes*

**fects****** Significant at 1 per cent. ** Significant at 5 per cent. * Significant at 10 per cent. Home: developed economies. Host: developing economies.*

**Note:***Pseudo R-squared is reported for PPML.*

**a**This econometric analysis shows that standard gravity models permit a meaningful explanation of FDi bilateral flows from developed to develop-ing countries. However, when the biTs variable is included, the results are ambivalent. Using one meth-odology (olS estimation of log-linear regression), results indicate that biTs have a positive impact on bilateral FDi, although the estimated magnitude of this impact is small. Since, according to recent lit-erature, this methodology produces biased estimates, an alternative method (PPMl) was also used. This method showed that biTs appear to have no effect on bilateral North-South FDi flows: the magnitude of

the estimated coefficients is close to zero. Moreover, the biT coefficients are not statistically significant;

in other words, results do not support the hypothesis that biTs foster bilateral FDi.

These results are consistent with the existing literature, which observes that the current state of the research is unable to fully explain the determinants of FDi, and, in particular, the effects of biTs on FDi. Thus developing-country policymakers should not assume that signing up to biTs will boost FDi.

indeed, they should remain cautious about any kind of recommendation to actively pursue biTs.