Andrew k. Rose and T.D. Stanley

Presented by:

María del Carmen Ramos Herrera.

.

Introduction

Meta- Analysis across studies

Publication Selection and Meta- Regression Analysis

Conclusions

My suggestions

.

Introduction:

The purpose of this paper is to use meta-analysis method to summarize, investigate and more accurately estimate the common-currency trade effect.

Meta-analysis can improve the assesment of this important economic parameter by combining all of the estimates, investigating the sensitivity of the overall estimate to variations in underlying assumptions, identifying and filtering out publication bias and later on use metaregression analysis( MRA).

This meta-analysis confirms a robust, economically important positive trade effect from monetary union.

.

The current interest in the trade effect of common currencies began with Rose (2000)

A panel of cross-country data covering bilateral trade between 186 different trading partners at 5-intervals between 1970 and 1990.

Since most of the variation is across pairs of countries rather time, Rose uses a “gravity model of trade”.

This resulting equation for assesing trade effects is the following:

.

Tijt : the natural logarithm of trade between countries i and j at time t.

β : set of nuisance coefficients

Dij: the log of the distance between i and j

Y: the log of real GDP

Z: other controls for bilateral trade

CUijt: dummy variable (currency union at t)

U: well-behaved disturbance term ∂ : partial effect of currency union on trade (ceteris paribus)

.

The surprising and interesting finding is that currency union seemed to have a very large effect on trade.

The coefficient for a currency union dummy variable has a point estimate of around 1.2 (Rose 2000).

This estimate implies that members of currency unions traded over three times as much as otherwise similar pairs of countries, ceteris paribus

(Why???) There was no previous benchmark in the literature, this estimate seemed implausibly large .

Almost all the subsequent research in this area has been motivated by the belief that currency union cannot reasonably be expected to triple trade.

.

Meta- Analysis across Studies:

It is a set of quantitive techniques for evaluating and combining empirical results from different studies.

Different point estimates of a given coefficient may be treated as individual observations.

Once compiled The hypothesis that the coefficient is 0

To estimate the coefficient of interest

more accurately

Rose and Stanley analyze 34 papers and 754 differents estimates of gamma.

There are a sufficient number of studies that have provided estimates of the effect of currency union trade and Meta-

Analysis seems an appropiate way to summarize the current state of the literature.

Most of them are representative and each estimate is weighted equally.

The central concern of this method is to test the null hypothesis: gamma = 0, where all estimates are combined.

The classic test comes from Fisher and this hypothesis is easily rejected at standard significance level.

However, Fisher’s test for overall effect is inappropiate for this and perhaps all areas of economic research.

Why???

Because: It is quite strict It is unlikely to be satisfied by empirical economics.

Then other tests for overall effects are needed. (Table 1).

We can see: fixed and random effects.

Manifestly, there is considerable heterogeneity across studies.

The fixed and random effects estimators differ greatly in magnitude and their confidence intervals don’t overlap

The smaller fixed effects estimate of gamma indicates currency union raises trade by 33%

The random effects estimate indicates this average effect is closer to 90%.

Note that all confidence bounds exceed zero

What does mean?

Positive trade effect

Table 2: Reports the fixed- effects estimates for gamma when studies are omitted from meta-analysis one by one.

There is little indication that any single studies is especially influential in driving this result.

Again, all confidence bounds are positive ( meaning positive trade efect from monetary union).

Another important isssue is that heterogeneity is present not only across studies, but also within most of the individual studies.

The random effect estimator is one way to accommodate heterogeneity. And MRA is another way to do it.

Publication Selection and Meta-Regression Analysis: (Critiques)

It is possible that these stong findings may be the artifact of selection for statistical significance (publication bias).

Publication Selection occurs when researchers, referees or editors have preference for statistically significant results.

Why???

Because insignificant findings tend to be suppressed.

The problem with such selection is that it will tend to exaggerate the magnitude of the empirical effect in question, potentially making negligible effects appear important.

Funnel graphs: (Important tool)

It is conventional method to identify publication selection.

It is a scatter diagram of precision (1/standard error(SE)) versus estimated effect.

In the absence of publication selection, the diagram should resemble an inverted funnel (wide at the bottom for small-sample studies, narrowing as it rises)

Asymmetry is the mark of publication bias.

Figure 2 show s lack of symmetry.

Funnel graph of 678 individual estimates:

To corroborate this pictographic identification of publication bias, we use an MRA of the t-value versus precision.

Publication bias is typically modelled as:

The reason behind this model of publication selection begins with the recognition that researchers will be forced to select larger effects when the standard error is also large.

Accounting for likely heterokedasticity leads to the weighted least squares (WLS) version of the previous equation:

In the absence of publication selection, beta o, will be zero.

In table 4:

Beta 0= 3.85, which is significantly positive, confirming the asymmetry of the funnel graph.

And we have a MODERATE corroboration of an authentically positive common currency effect (t=1.97)

Then, after accommodating publication bias, an economically significant trade effect of monetary union remains.

Results: A 95% confidence interval for gamma, after correcting for publication bias, is 0.184-0.586 Trade is increased between 20 and 80%.

Conclusions:

First, the hypothesis that there is no trade effect from currency union is robustly rejected when individual studies are pooled.

Second, the pooled estimate is not only positive, but also economically significant

Third, there is evidence of publication bias and after correcting this problem we will have a lower trade effect from monetary union.

Fourth, as expected, a number of research characteristics are found to have a significant effect on the reported common currency effect.

Meta-analysis has an important limitation

If there is a common, systematic bias across the entire literature, meta-analysis has no way to distinguish it from an a unique empirical effect.

My suggestions:

First, to clarify Rose and Stanley question:

How much would trade increase when countries have a common currency?(In this case, these countries have always shared it)

Second, I wonder another question:

How much would trade increase if I introduce a new common currency in these countries ?(But in this case, these countries have never shared it)

Finally, we could do this analysis in a dynamic way (paying attention to the long term) To see the effect on trade across time when currencies are created and destroyed.

QUESTIONS AND COMMENTS

THANK YOU VERY MUCH FOR YOUR ATTENTION