The new twist that Kwak offers is retrospective wealth taxation, in lieu of retrospective income (or capital gains) taxation. While I don't think the proposal is likely to end up having political traction, it's a great topic that can be very illuminating in thinking about income and wealth taxation design issues. Kwak, who does not have an extensive tax practice or teaching / scholarly background, has nonetheless done an impressive job of mastering relevant aspects of the very complicated tax policy literature concerning income and consumption taxation.
To give a sense of what retrospective income and wealth taxation are all about, a picture is worth n words (where n ≥ 1,000). Let's start in this post with retrospective income taxation, as developed by Auerbach and Bradford in their work. I will then turn to the retrospective wealth tax in my next post.
Sorry if this chart seems hard to read - this problem should fix itself if you click on it. To put the issue in perspective, suppose we know three things about an asset that the taxpayer just sold: the amount realized on the sale, how long she held the asset, and what she initially paid for it. A realization-based income tax discourages selling appreciated assets (and encourages selling those that have lost value) due to deferral of the unrealized value change. And the problem gets exponentially worse if, as under the current U.S. income tax, the tax will be permanently eliminated if one holds the asset until death.
A logical response to this problem might be deeming the gain (or loss) to have accrued ratably (with compounding) over the asset's holding period. Then interest might be charged on the deferral, based on the extra tax that would have been due in prior years by reason of current inclusion. But here's the problem. Under constant information, assets might be expected to appreciate at just the risk-adjusted normal rate of return. An asset that did better than this must have jumped up or down in value when information about it changed (e.g., a risk was resolved favorably).
Let's simplify the story to allow for just one relevant information change, and put it in terms of the above chart, where we know that the asset was sold for the price and at the time shown at the upper right. The assumed value path shows how much it "must" have been purchased for if there was never an information change. But suppose it was purchased for less (as in the case of "Actual (Winner)" or for more (as per "Actual (Loser)." Then it must have appreciated from that starting point until the information shock occurred, whereupon it jumped to the Assumed Value line and stayed there the rest of the way.
Suppose we add the following assumptions. First, the taxpayer always knows the value, but the government can only observe the purchase and sale transactions (time and price). Second, the taxpayer cannot predict the time or direction of the information change that leads to the up-or-down value jump. Or more precisely, information is symmetric - she can't outguess the market, which has incorporated the already-known information into the asset price.
Having the tax system assume ratable accrual, as determined at the time of sale, would amount to drawing a straight line from the actual purchase price to the eventual sales price, rather than having the correct slope (at a higher or lower level) followed by adherence to the assumed path. Thus, suppose that, in the "Winner" case, the taxpayer knows that there has been a favorable value jump. Even if we were allowing for further, as yet unpredictable, value jumps, the taxpayer would now expect the asset, on average, to stay on this path. But note that, once the (positive) value jump has occurred, the assumed value path under ratable accrual always treats value as lower than it actually was. Only when the sale occurs does the assumed value path converge with actual value. So, the longer you delay the sale, the longer you get to "average forward in time" the amount of the value jump.
This creates lock-in for an appreciated asset that has had a positive value jump. So it fails to eliminate lock-in, although it does reduce the magnitude of the problem.
Suppose one wants to eliminate that lock-in problem altogether. As Auerbach and Bradford realized and explained, the easiest way to do it is by simply ignoring the actual basis of the asset in the taxpayer's hands. Instead, the tax system would simply use the amount realized upon sale, and the holding period, to impute gain based purely on the Assumed Value Path. In effect, one would use a fictional basis in lieu of the actual one. Then one would determine the gain that "should" have been taxed in earlier periods, compute the effect on past years' tax liabilities, and make the tax adjustments payable at sale with requisite interest.
This seems odd - what might motivate it? Now again, it truly does eliminate any tax-induced incentive to hold or sell the asset at any time, under the assumption that the taxpayer can never outguess the assumed value path suggested by the value that is known to her at any particular time. But don't we also actually want to get the amount of the tax right?
Under certain assumptions, NO. Suppose we think of the gain or loss, relative to the assumed value path, as purely reflecting risk. And suppose further that we believe taxpayers will adjust their investment portfolios, in light of the tax, to get to the exact point along the risk-return frontier that they like. For example, if gains are taxable and losses are refundable, that amounts to making the taxpayer's portfolio less risky (and with a lower expected return) than she preferred. So in theory she can respond by selecting a riskier pre-tax portfolio and getting right back to the same place.
Not so fast? Well, this is a common assumption in models, albeit not necessarily to be found in the real world. It requires complete markets, consistent rational choice, a flat tax rate including loss refundability at that rate, and the ability to borrow at the risk-free rate.
To the extent that it is true, however - i.e., assuming a can-opener - ignoring actual basis just doesn't matter. It merely reflects a risky outcome that the taxpayer can get to with or without the tax. (Also, with complete markets, rational choice and no externalities, there is no reason to offer the tax system's insurance feature by taxing gains and reimbursing losses - that is for cases where private insurance is crippled by adverse selection, e.g., with respect to the "ability lottery" and under-diversified human capital).
Accordingly, under sufficiently rarefied assumptions, retrospective income taxation makes sense as a way of eliminating distortions regarding when to sell assets, without any relevant downside given the assumptions that relate to issues of risk.
Even within this rarefied (and indeed, concededly unrealistic) scenario, there is an important limiting factor to keep in mind with regard to allowing the use of this method. It is appropriate SOLELY with regard to what David Bradford, in designing his "Blueprints" cash-flow consumption tax system in the 1970s, called "arm's length" transactions. Within Bradford's meaning, if I purchase Facebook stock at its current market price, that is an arm's length transaction. But if Mark Zuckerberg does so, be it at the pre-IPO stage or even today, if he can affect its value (e.g., by supplying labor that he uses to increase the stock's value, because he is not paid a wage equal to the value he has contributed), that is non-arm's length.
In Bradford's Blueprints cash flow consumption tax system, yield-exempt treatment of investments was allowed only for those that qualified as arm's length. Expensing had to be used for those that were non-arm's length, so that the tax system would reach inframarginal or labor-related returns.
In the context of retrospective income taxation, allowing use of the Assumed Value Path for non-arm's length transactions, such as Zuckerberg's getting Facebook stock when he was creating the company, would be unacceptable. It would not even avoid tax-discouraging sale if he anticipated being able to create further value jumps through further contributions of under-priced labor in the future (assuming that, post-sale, he could no longer avoid being taxed on such contributions, e.g., because he would now have reason to demand a full-value wage). But even if it did permit the timing of sale to be tax-neutral, it would be unacceptable because we want to tax people on the economic value that they derive through their "ability" plus labor supply.
Summing up the merits, if (and insofar as) we buy into the full risk, etcetera, analysis, Auerbach-Bradford retrospective income taxation may be an appealing method for the taxation of asset sales that are not taken into account, for tax purposes, until the gain or loss is realized. But it would have to be limited to arm's length transactions in order to be acceptable. And it may require taking a couple of stiff drinks, so that one is ready to accept the risk analysis and thus to ignore actual cost basis with no tears.
Even so, it seems likely to be politically unacceptable. Returning to Figure 1, it taxes a gain to someone who actually suffered a loss. (Please indulgently note that I meant to show the actual cost basis for the loser as lying above the eventual sales price.) That is likely to be optically unacceptable, even if you yourself have had enough of the spiked kool-aid to accept it with equanimity.
Plus, relative to an income tax based on correct actual values that were taken into account annually, it under-taxes the big winner who earned more than the normal rate of return.
So it looks like anti-insurance: under-taxing winners, and not just over-taxing losers but imputing fictional gain to them. This is not something that one is likely to be able to "sell" to policymakers, even under a far less politically fraught environment than one could imagine us ever actually facing.
The retrospective income tax has therefore never really gotten anywhere, even in the tax policy community. My sense, which I believe others share, is that it's a clever idea from which we learn something - it helps us to think more deeply about real world effects and their causes, and about the universe of proposals that might actually be politically feasible at some imaginable point. But there has been relatively little work done towards showing how it might actually be implemented, because the effort seems too unlikely to bear fruit.
But here's the thing about retrospective wealth taxation, as proposed by Kwak (and as I will discuss in my next blog entry). Despite the strong analytical overlap, it's surprisingly different, and in some respects has opposite effects on who wins and loses from its being used in place of an annual wealth tax based on actual (albeit, to the tax system unknowable) values. This adds to the interest of analyzing and writing about it, even if one does not end up concluding that it might actually be added to the universe (or shelf) of potentially feasible proposals.