This blog sucks, but the heart is in the right place.
I don't want to make a habit of sharing my opinions on a blog dedicated to data collection and dissemination, but this blog blows.
I've decided to critique my own project because everything about it is widely unpopular. Why would it be popular? This blog only appeals to someone like myself, and even I am not satisfied with it. There are numerous critiques I could launch at my own work.
Reasons why this project has raised 0 USD in four months on GoFundMe:
1. I am not a homeless veteran; I am not making a liberal documentary; my target audience is tiny and largely self-absorbed.
2. People don't think it is important to have more historical and disaggregated datasets.
3. Economists and Investment Researchers are not interested in long financial and economic datasets
4. Donating is not rational. Why donate when I've already made many datasets available at no cost?
5. Lack of publicity, and formal documentation. Formal documentation would draw attention to the project.
6. The datasets are not available on Quandl, FRED, or Bloomberg, and are not automatically updated each month.
I began with my assessment of the blogs failure. There is a clear path forward. Money would help.
Now I must change my tune and discuss why long data-sets may remain unpopular even with greater convenience, publicity, and documentation.
Economists generally stick to using short-samples in empirical work. The main reason they often give is "availability", but another reason not vocalized is the petrifying anxiety they feel when faced with large shocks and structural changes. The economic data may actually exist in some little known government or trade publication, but the published data is often very short. I think most of the collectors of such data would argue that going back in time is futile because of 'dramatic' changes in definitions and survey questions yata yata. Although structural changes are certainly inconvenient, it is unwise to shorten an empirical sample because of an easily remedied level, or trend shift. A structural break caused by a definitional subtraction or addition can be accounted for by utilizing impulse, step, or trend indicator saturation.
If one is to even have a chance of drawing the correct inference in cointegration analysis then accounting for structural changes is a necessary condition. I am not not just referring to structural breaks caused by measurement but also those caused by the major changes that have occured from time to time. You know the changes. Gold Standard this, Cival war that, inflation expectations this, demographics that, financialization this, financial innovation that.
Economic growth is nonstationary. Economists should be more empirically comfortable with structural changes than they currently are. Why do forecasts fail and models fail? The most common explanation is the dastardly level shift.
When a macro economist intentionally estimates a model over a period in which bond yields are falling an Angel loses its wings. If the goal is to empirically confirm a theory we told ourselves then mission accomplished. The theory might be flipped on its head if we considered only a period of rising bond yields.
People are lazy, they look around and proclaim "Well if this is the way things are now, then this is how they will always be". The forecast is for no change - safe move. They are also inclined to believe that people are dramatically different today than they used to be - they think that naturally we must be smarter with quantum technology. There is this tendency to assume that people 100 years ago were apes compared to modern man. 100 years ago they had the same short- ass datasets market players are using today. We continue to act like apes.
I will continue this rant in my next post.