Adding Zero values for missing trade to balance the panel variable data

[James Baloto]

Hello everyone,

I'm analyzing bilateral trade exports over 20 years from 5 exporting countries to all importer countries, using panel data with fixed/random effects models and PPML estimation. The data is unbalanced, tracking trade per commodity for each exporter-importer pair. I'm considering imputing zero values for missing trade activities to reflect the absence of trade in certain years. Would this approach be appropriate, or could it introduce biases? If i will add zero, is there a way to fill each commodity/exporter-importer/year with 0? Any thoughts or suggestions on this?

Thank you!

This is what my data look like:

year	exp	imp	exp_id	imp_id	contig	lang	dist	procde	prodesc	trade	o_wgi	d_wgi	forest	gdp_exp	gdp_imp	unit_level
2002	IDN	AFG	80	2	0	0	6001	4409	Wood	10317	- 1.10	- 0.53	1009558	427825987584	7555180544	2
2002	IDN	AFG	80	2	0	0	6001	4414	Wooden frames	14811	- 1.10	- 0.53	1009558	427825987584	7555180544	2
2002	IDN	AFG	80	2	0	0	6001	4420	Wood marquetry and inlaid wood	15274	- 1.10	- 0.53	1009558	427825987584	7555180544	3

[Clyde Schechter]

I'm considering imputing zero values for missing trade activities to reflect the absence of trade in certain years.

First a disclaimer: I'm an epidemiologist, not an economist or econometrician. So there may be considerations of trade economics that I'm not aware of that might modify my advice.

But from a general statistical perspective, the question is, do the missing trade activities really represent an absence of trade? Or are they just an absence of information about trade? If they really represent an absence of trade, then not only will it not introduce bias to replace those by zeroes, it will reduce the bias that your data set starts out with by virtue of not including those zero-observations. In other words, filling in those gaps with zeroes is, in my view, mandatory.

But if the absence of some trade activities in the data set represents simply the inability to report the trade that took place, for whatever reasons, then filling them in with zero would likely introduce a very severe bias into your data set.

So I think your first, and very important, task is to become certain about why those years have no trade reported: is it really because there was no trade to report, or was the trade there but it went somehow unreported?

[Jeff Wooldridge]

I'm with Clyde. The important issue is whether the missing value is due to someone not wanting to insert a zero to indicate no trade, or whether it's actually a missing value. As Clyde said, replacing actual missing data with zero can have very bad effects.

[James Baloto]

Thank you for your insights and recommendations. I will check again with UNCOMTRADE database to determine whether the issue is due to missing data on certain trade activities or an actual absence of trade.

[Joao Santos Silva]

Dear James Baloto,

For what is worth, my understanding is that the data records reported trade, so if there is no trade, none is reported and zero trade is therefore coded as missing. Of course, very small values of trade are also not reported and these are also recorded as missing. The standard practice is to also code these observations as zeros because it is expected that this small rounding error will have little effect on the results. As an aside, that is possible the least of our worries working with trade data, e.g., large discrepancies between the values reported by importers and exporters are more concerning to me.

Best wishes,

Joao

[James Baloto]

Thanks for explaining how does the trade data is handled and especially on the standard practice about zero and small trades. I appreciate your insights!