I've been doing a lot of reading on taxes and public spending recently. I had enough of an overview of the subject that this hasn't made me really rethink anything I've written about taxes before, I'm still for relatively balanced forms of taxation between consumption, capital, and labor. I have changed my view on the incidence of taxation a bit, because of high levels of tax evasion I've had to revise my view that taxing at the individual level is optimal, rather some taxation of corporate profits is necessary to mitigate tax avoidance. I've also come to favor taxing property relative to income even more highly, but I already tended this way.
But, while my thoughts on the general themes haven't changed much, there were a few real revelations on more specific, particularly methodological aspects.
First of all, I was a bit shocked how sensitive most of this literature is to assumptions. In any social science there does tend to be some sensitivity to this, but in this literature there are sign changes and not just shifts in relative valuation.
There are a lot of what I can only call "well, duh" instances in this literature. For instance, a lot of the basic parsimonious models find that not taxing capital is optimal. However, add in terms for human capital or make government spending at least partially a complement to private investment and all of a sudden it becomes optimal to tax capital. Duh. If your only source of growth is capital investment and it's simply a choice between capital and consumption then of course you don't want to tax capital, it lets you consume more later. But, if labor also creates value and not just capital then you don't want to distort the capital labor trade-off, this becomes inefficient. Similarly, if government does something of value then there is a trade off here and it is optimal to tax capital (or if externalities exist terms change, leisure trade offs matter, time sensitivity and intergenerational responses matter, etc.). That anyone would take the basic model as a policy prescription is dumbfounding. Obviously policy requires the extensions of the model that take real-world complications into account. Keep this in mind when you come across something in the paper vaguely alluding to an economist or a given model, these models are very highly sensitive to starting assumptions and many of them give the unsurprising result that if you assume something to have no value and that something else does have value, your model will confirm this. Surprise, surprise.
Empirical studies tend to give highly mixed results, this makes me tend to believe that we don't have a good handle on what is optimal. If empirical results give different answers based on selection and time period choices or based on whether or not human capital terms are used, then we probably aren't at a point where we can make optimal policy recommendations. This may not be because of a weakness in the theories, there may have been too much economic changes over the past two centuries and longer time frames are necessary.
Another takeaway is that there should be some scepticism regarding results that reach into infinity. Some models have negative welfare impacts for decades and centuries on the way to optimum paths. This is obviously nonsensical as a policy approach, there are too many potential catastrophic downsides over this long of a time period for it to be rational to think in this frame, even if perfectly altruistic. While a small welfare loss may be justifiable for a large increase, any large loss simply isn't justifiable without a great deal more certainty than we have. Few states survive for more than a few centuries and their collapse tends to bring down the fortunes of most of its citizens, this is not a sensible time frame to use for anything.
The main takeaway though is that when it comes to taxes at best we have a general sense of the tradeoffs involved. Anyone that claims to have some idea of what an ideal rate or mix of taxes would be is full of shit. All of these approaches are highly sensitive to starting assumptions and methodologies and the empirical modeling gives answers that are far from robust. Valuable knowledge is being gained but we're far from where a technocrat can give a valid plan for society. What has gotten better is that we can understand the menu better and what we're gaining and losing when we commit to a choice. Doesn't mean that we'll necessarily like the dish served, but at least we'll have only ourselves to blame, rather than fortune, which was the only recourse of the past.