The question I have is regarding the formulation of the initial hipotheses for testing. But before that, I need to choose and define my methodology approach. I am studying first year graduate economics. The subject of my thesis is measuring the impact of labour and capital inputs on production and income at the national level (or industrial level).
Now, as far as theoretical basis goes, most of my uni professors are either neoclassical, or ex-planned economy-recently-turned-neoclassical. I can't say for sure, because I don't know them that well yet, of course, but based from what I've seen, for example, on the scientific databases that are available for journal reading and study of scientific literature, I haven't found any post-keynesian ones (journals) subscribed by my uni. That's not a problem in itself, as I need to be aquainted with the mainstream methodology and theory first, before I can compare methodologies or theory. (The reason I got a bit distracted by the theoretical side is I got a bit confused about my topic. As I'm going to measure input impact on production AND income, isn't my topic actually supply and demand analysis? Well, I think it is in a big way. And as such, I really think the post-keynesian school could really tell a lot about the demand side. But there is no way for me to access post-keynesian economic journals...)
However, my question is not regarding theory, but methodology. As I need to measure the impact of capital and labour inputs on production, I've so far found three large sets of methodologies that can be used: 1) Growth accounting. 2) Aggregate production function analysis. 3) Econometric modelling.
As far as Growth accounting (1) goes, it's basically using the Solow form of Cobb-Douglas production function for measuring relative factor shares, changes in factor shares and changes in TFP (total factor productivity). EUKLEMS and WIOT are databases used for these sorts of measures.
As for APFs (2), I've read so much about various forms of aggregate production functions, but I'm still not sure how to estimate them. The basic idea (due to my weak prior econometric training, or lack-thereoff (my own fault)), if I'm correct, is you take time-series data of some form of inputs (i.e. total hours worked per year for labour input and GFCF for capital input), you use the linear least squares method on the inputs AND the output, and then, I guess, you find the production function parameters? I.e. to make that much of Y you need this much L and that much K. Okay. But If I want to test the production function econometrically, I need a linear model of it. So I use logarithms to obtain the linearized form. But which step is first? Do I first estimate the parameters and then linearize, or do I linearize to more easily find the parameters? I might have gotten this confused and backwards... Now, I might as well use CES and/or Translog production functions. (Please, correct me if I'm wrong, but isn't the Translog production function just the same CES, only a special case of it (the Cobb-Douglas case)?
And as for the (3) econometric analysis, as strange as it may sound to myself, this method looks the most straightforward to me. I can make a VAR (vectoral autoregression) model using GDP or GO (gross output), or GVA (gross value added) as my dependent variable and my Labour (Employees or employees+self employed persons; hours worked by employees or hours worked by employees+self employed persons) and Capital (Gross fixed capital formation or "Capital services" (still not sure how this is measured)) inputs. I can make a structural VAR, and I can even make a VECM (accounting for the levels, i.e. incorporating the non-stationary I(1) data as well as stationary I(0) data into the model). Then I would analyze the impulse-response functions and make inference that way. (I've done an initial testing of the data, and it might be so that the labour input series is of different integration order than the output (GVA) and Capital time series data. I don't know how or why. But if it were such a case, I would need to use an ARDL model? Suggestions welcome, please :))
So, I guess, after all this writing, my questions are... How do I choose the right methodology? I want to use them all, because they seem similar and connected, and it looks as though I could make comparisons, inference, etc. But I'm actually overwhelmed by them right now. I need to submit my initial paper real soon, and I don't want to miss out on something important if I discard any one of these methodologies. But on the other hand, to use them all would be just too much for me right now. I've actually had little formal Macroeconomic training, as my bachelors was in BBA, so you can see why I'm not so confident in choosing the right method to accompany the theory.
Many thanks in advance R.