Dear all,
I am currently working on a project where I want to predict the size of the manufacturing sector depending on the exposure to trade with China. Because I am concerned about endogeneity I want to instrument for my explanatory variable. I am dealing with around 40 developing countries, which all import from China (IPWitc). IPWitc are the imports for country i at time t from China (c). I would like to instrument the variable IPWit with the imports for developed countries from China. I have trade data on 8 developed countries, meaning I have 8 instruments: IPW1tc, IPW2tc, .., IPWntc. Note that technically these are all different variables, so technically I suppose I could also remove the individual identifier i.
Here is where I am not sure if I have a problem when estimating my model. IPWitc is being instrumented for every developing country by one (the same) developed country for all instruments. E.g. the imports from China to the US are being used once for every individual identifier (every developing country). Would I need at least 40 developed countries (and then randomly assign country pairs) in order to instrument for the 40 developing countries I am dealing with?
In any case, my Sargan J statistic has a p-value that is smaller than 0.0000, so I am sceptical about my instruments.
Does anyone have any advice?
Best,
Hylke Dijkstra
I am currently working on a project where I want to predict the size of the manufacturing sector depending on the exposure to trade with China. Because I am concerned about endogeneity I want to instrument for my explanatory variable. I am dealing with around 40 developing countries, which all import from China (IPWitc). IPWitc are the imports for country i at time t from China (c). I would like to instrument the variable IPWit with the imports for developed countries from China. I have trade data on 8 developed countries, meaning I have 8 instruments: IPW1tc, IPW2tc, .., IPWntc. Note that technically these are all different variables, so technically I suppose I could also remove the individual identifier i.
Here is where I am not sure if I have a problem when estimating my model. IPWitc is being instrumented for every developing country by one (the same) developed country for all instruments. E.g. the imports from China to the US are being used once for every individual identifier (every developing country). Would I need at least 40 developed countries (and then randomly assign country pairs) in order to instrument for the 40 developing countries I am dealing with?
In any case, my Sargan J statistic has a p-value that is smaller than 0.0000, so I am sceptical about my instruments.
Does anyone have any advice?
Best,
Hylke Dijkstra
