Hello all!
I am trying to find out what are the Stock-Yogo threshold values for the case of 6 endogenous variables? Finding these critical values seems straightforward for the cases of one or two endogenous variables. However, I am struggling very much for cases with higher numbers of endogenous variables (e.g., the case of a model with six endogenous variables). Let me elaborate further below:
My aim is to find out whether my instruments are strong enough, by comparing the Kleibergen-Paap rk Wald F statistic and the Cragg-Donald Wald F statistic to the Stock-Yogo (2005) threshold values.
I am running this analysis as part of a paper where I study the relationship between a firm strategy (STRATEGY), my IV of interest, and a firm’s performance (PERF), my DV. Further, the paper studies a total of five interaction effects. I have a 2SLS model, drawing on panel data with fixed effects.
As a next step, to get the Kleibergen-Paap rk Wald F statistic, the Cragg-Donald Wald F statistic, as well as the Stock-Yogo (2005) threshold values, I run:
Luckily, even for my full model with the 6 endogenous variables, the xtoverid command does give me both Kleibergen-Paap and Cragg-Donald statistics. However, for my full model with the 6 endogenous variables, xtoverid sadly does not give me the Stock-Yogo weak ID test critical values. The output window in Stata merely shows the following:
Stock-Yogo weak ID test critical values: <not available>
(This is different from my models where I have only one or two endogenous variables: Here, xtoverid correctly computes all three values: Kleibergen-Paap, Cragg-Donald, but also Stock-Yogo 2005 threshold values)
Even more sadly, I also cannot find the threshold values for the case of 6 endogenous variables anywhere in the literature. The key reference for Stock and Yogo (2005) is the below. The four relevant tables are on pp. 58-61 of Stock and Yogo (2005). Table 1 (approach based on the bias method --> TSLS Bias) only shows the threshold values for cases of up to three endogenous variables (n=3). Similarly, Table 2 (based on the size method --> TSLS Size) only shows the threshold values for cases of up to two endogenous variables (n=2).
Am I missing something? Do I misunderstand something?
Do you have any advice on what the threshold values are, or how I could compute them myself?
I am massively grateful for any advice on this!
Franz
I am trying to find out what are the Stock-Yogo threshold values for the case of 6 endogenous variables? Finding these critical values seems straightforward for the cases of one or two endogenous variables. However, I am struggling very much for cases with higher numbers of endogenous variables (e.g., the case of a model with six endogenous variables). Let me elaborate further below:
My aim is to find out whether my instruments are strong enough, by comparing the Kleibergen-Paap rk Wald F statistic and the Cragg-Donald Wald F statistic to the Stock-Yogo (2005) threshold values.
I am running this analysis as part of a paper where I study the relationship between a firm strategy (STRATEGY), my IV of interest, and a firm’s performance (PERF), my DV. Further, the paper studies a total of five interaction effects. I have a 2SLS model, drawing on panel data with fixed effects.
- Things work fine for the model where I only look at my IV of interest (i.e., where I have one endogenous variable). Things also work fine for the models where I look at each of my interaction effects individually (i.e., models in which I have two endogenous variables: the IV of interest, and the respective interaction effect). For all these models, xtoverid command correctly tells me the Stock-Yogo threshold values.
- However, this sadly does not work for my full model, i.e., the model where I look at my IV of interest, but also include all of my five interaction effects at the same time. In my full model, I have a total six endogenous variables (main IV of interest + the 5 interaction effects). Thus, I need to use a total of six instruments. For my full model, I use the below commands:
Code:
xtivreg PERF Control1 Control2 Control3 Control4 Control5 Control6 Control7 Control8 Control9 Control10 Control11 (STRATEGY Interaction1 Interaction2 Interaction3 Interaction4 Interaction5 = Instrument1 Instrument2 Instrument3 Instrument4 Instrument5 Instrument6), fe vce(cluster FirmID)
Code:
xtoverid, noi
Luckily, even for my full model with the 6 endogenous variables, the xtoverid command does give me both Kleibergen-Paap and Cragg-Donald statistics. However, for my full model with the 6 endogenous variables, xtoverid sadly does not give me the Stock-Yogo weak ID test critical values. The output window in Stata merely shows the following:
Stock-Yogo weak ID test critical values: <not available>
(This is different from my models where I have only one or two endogenous variables: Here, xtoverid correctly computes all three values: Kleibergen-Paap, Cragg-Donald, but also Stock-Yogo 2005 threshold values)
Even more sadly, I also cannot find the threshold values for the case of 6 endogenous variables anywhere in the literature. The key reference for Stock and Yogo (2005) is the below. The four relevant tables are on pp. 58-61 of Stock and Yogo (2005). Table 1 (approach based on the bias method --> TSLS Bias) only shows the threshold values for cases of up to three endogenous variables (n=3). Similarly, Table 2 (based on the size method --> TSLS Size) only shows the threshold values for cases of up to two endogenous variables (n=2).
- Stock JH, Yogo M. 2005. Testing for weak instruments in linear IV regression. In Identification and inference for econometric models: Essays in honor of Thomas Rothenberg, Andrews DWK, Stock JH (eds). Cambridge University Press: NY: 80–108.
- The tables can also be found on pp. 39-42 in the earlier 2003 version of the article: https://scholar.harvard.edu/files/st...regression.pdf
Am I missing something? Do I misunderstand something?
Do you have any advice on what the threshold values are, or how I could compute them myself?
I am massively grateful for any advice on this!
Franz
