What does "endogeneity of the industry structure" means?

Dong,2019,page 899 documented that:

We estimate Export Market Leniency Laws as the weighted average of the passage of leniency programs in all other countries, excluding the country in which the firm is headquartered:

$$\text{(Export Market Leniency Laws)}_{jkt} = \sum_k w_{kj}L_{kt},$$

where $$k$$ denotes any country other than country $$k$$, $$j$$ denotes a three-digit SIC industry, and $$t$$ denotes the year. $$w_{kj}$$ is the share of the three-digit SIC industry $$j$$’s exports from country $$k$$ to any other country $$k$$ out of all of the exports from industry $$j$$ in country $$k$$ in 1990. $$L_{kt}$$ is an indicator variable that takes a value of one if country $$k$$ passed a leniency program by year $$t$$, and zero otherwise. To avoid any endogeneity of the industry structure, we base the weights on the data in year 1990. The variable ranges from zero when none of the foreign countries that receive any exports from the firm’s industry have passed a law to one when all of the foreign countries with exports from the firm’s industry have passed leniency programs.

I do not understand what does "endogeneity of the industry structure" means in this context, and why we need to base the weights on the data in year 1990 to avoid this issue?

• 1. writing out equations is not done not to annoy people but in order to make the equations searchable. Please have a look at this link and familiarize yourself with mathjax math.meta.stackexchange.com/questions/5020/…. Since you are student of economics it will do you good for a future since most respected academicians use latex for their research and latex equations work virtually identically to mathjax 2. You should not post pictures of a text.
– 1muflon1
May 8 '21 at 8:45
• This time I edited it for you please next time try to do it yourself at least for the text
– 1muflon1
May 8 '21 at 8:53

Broadly speaking endogeneity means that something some variable is determined within model as opposed to outside it. More narrowly in econometrics it means correlated with the error term (Wooldridge, (2009). Introductory Econometrics: 4ed. p. 88). These meanings are related since one will often lead to the another.

In this paper they actually use it in both meanings.

1. From economic literature we know that industry structures (e.g. whether industry is monopoly, duopoly, oligopoly, monopolistic etc.) depends on institutional setting and laws as well (e.g. see discussions of this in Belleflmame & Peitz Industrial Organization Part IV). Moreover, it is often further argued that in turn laws that country implements also depend on "economic conditions of a firm’s country" (Dong, 2019). So what they mean is that industry structure will not be just given by some exogenous parameters such as let's say availability of local natural resources (this could be an exogenous force that could determine if industry is a monopoly e.g. if there is let's say only 1 viable diamond mine then this would be exogenous factor leading determining market structure to be a monopoly). Rather, here the market structure is shaped by the leniency programs that affect cost of collusion and in turn the leniency programs likely depend on economic conditions (e.g. a country with less problem with cartels might have less anti-trust laws).
2. Dong (2019) throughout the chapter in previous paragraphs Dong repeatedly states that they are dealing with endogeneity because it could be a valid critique of identification strategy (i.e. their empirical model). Consequently, they also mean it in more narrow way that it would mean that in their regression of margins on those leniency laws the leniency laws would be correlated with error term. In fact, this is why Dong says that they dealt with it by using leniency index from 1990s - using past ('lagged') values instead of contemporaneous one's is a common econometric technique to deal with endogeneity in an empirical model.
• @PhucNguyen no competition or amount of collusion are economic conditions. Also GDP is an indicator (although it too indicates economic conditions).
– 1muflon1
May 9 '21 at 11:35
• @PhucNguyen using lagged variables is most basic recommendation you will find in most econometric textbooks (although it does not work always). Papers usually do not go into great detail into this because it was already known for a very long time. For ways how to deal with endogeneity you can have a look at verbeek a guide to modern econometrics chapter 5. Regarding justification for lagged variables as instruments in macro context you can have look at Romer Advanced Macroeconomics pp 376. Most important feature of instrument is that it is uncorrelated with an error term. Anything that
– 1muflon1
May 9 '21 at 11:53
• is presently known and occurred in past should not be correlated with errors if people are rational and forward looking (although, it is best to still test instruments because even though we know people are not completely myopic behavioral econ shows that sometimes they are not fully rational and in some context that might mean that errors today will depend on past realization of variable you are using).
– 1muflon1
May 9 '21 at 11:54
• @PhucNguyen I don’t know enough about the paper or the topic to answer that, my first guess would be it’s because of some data issue.
– 1muflon1
May 9 '21 at 22:56
• @BeautifulMindset if it does not have time subscript it should be time invariant. Regarding that statement it’s not a direct quote but I am paraphrasing (direct quotes are in “”) in the Dong 2019 they state that “passage of leniency programs might not be fully exogenous to the political and economic conditions” they state that on pp 15 and literally provide you with further references there. If that would not be case then that sentence in the paper would make no sense. But, worrying about endogeneity of industry structure is common in IO, you can see the textbook I quoted above.
– 1muflon1
Jun 16 '21 at 21:16