Working papers that (shamefully) are not yet published

The reviewing is, clearly, with lots of noise. Unfortunately, the noise can imply that some (very) good papers do not appear to show up in major journals. Why is this? Is it that the author(s) give up after trying a few top journals? Is it because it is the same draws of reviewers? I'll note here a few papers that 'have been around' for some time, are often well-cited, and most certainly deserve a good look, whether published or not. 

Important disclaimer: These are just 'samples' that have come to my attention; I do not claim to know all good papers that have not yet gone on print, nor do I systematically try to include a paper I see and like in this list unless I feel it has been around for some time.

This paper is the winner of the AAA Best dissertation award, and a very well-known paper in Banking research. It asks the question as to whether we observe contagion due to Fair Value accounting during the crisis, by looking at extreme negative bank returns, and finds very clear evidence consistent with the models of  Plantin, Sapra and Shin (2008) and Allen and Carletti (2008). My reading: Fair-value accounting may be small in magnitude (Laux and Leuz, 2009) but it seems to carry empirically a large informational role that contributes to stock crashes.

This is the first paper that carefully derives the type of interference to loan signalling that is caused by Mark-to-Market (MTM), and is one of the first papers demonstrating the idea that using market prices for pricing inside loans can, in the end, shut down the flow of information to the market. To be more precise, it is more difficult to signal quality under MTM because the bad type gets an extra benefit from 'fooling' the market than was not present under historical cost. Of note, the paper is referenced as a relevant study in G. Plantin and J. Tirole's recent paper, no small endorsement.

The study challenges the conventional approach to detecting earnings management by starting from first principles, and solving a manager to optimally engage in earnings management over time - to be matched to data. Managers in this model look at both current and future consequences of their actions. The results: earnings management may be much more widespread than we think, with about 3 out of 4 having manipulated at least once during their tenure. I'll refer to this Harvard forum for an excellent overview of her work and results.

Regression-based Earnings Forecasts, by J. Gerakos and R. Gramacy
This is an apparently simple idea: how do we forecast annual earnings, what statistical models perform best? But we need actually this statistical exercise for many studies that use 'expected' earnings as an input. There is actually a well-regarded history in economics that shows that random walk models appear to dominate sophisticated statistics for many problems. Gerakos and Gramacy are the first to show this claim rigorously for earnings, and this work (which is a lot less easy as it would seem) should be a reference for a lot of literature.

This is a risky paper because it offers some empirics with a plausible theory that is not yet made completely formal; but the theory is now worked out in a recent paper (here) and, together, the studies give new meaning to the problem of complementarities between information sources. This provides a contribution to prior explanations that rely on investor sophistication (which is another plausible, but conceptually different, explanation).

The paper is a living proof that very senior people do not necessarily have an easy ride publishing good papers. The idea of the paper is simple, let us discuss the outcome of a voluntary disclosure problem in which the details of the litigation environment are explicitly modeled (over the usual assumption of taking perfect enforcement of truthful disclosure as a given). The interested reader may also check an excellent related paper that speaks to this issue within a different setting (here).


Congratulations to papers removed from the list!

This paper explores a way to model information disclosure by way of a reduction in Shannon's entropy (namely, how dispersed posteriors can be). There is a technological lower bound on entropy, so what is the best information system? In a signalling game, the surplus is a function of the surplus of the lowest type, so the optimal information system should always be impairment-like and identify the lower types. A very robust statements within the signalling paradigm.
now forthcoming in Journal of Accounting and Economics

The Effects of Real Earnings Management on the Firm, its Competitors and Subsequent Reporting Periods, by C. Chapman
This paper is a serious attempt to document evidence about real earnings management based on detailed micro-data, namely the author is sometimes known as the 'soup' person because, instead of relying on large-sample proxies, he has used detailed data about soup discounts. Surprisingly, there seems to be resistance about an approach in which we understand completely what the proxy is. 
published in extended form in Journal of Accounting Research, Real Earnings Management in Sales, with M. Ahearne, J. Boichuk and T. Steenburgh.

What happens if firms compete on incentive contracts for the best CEOs? The answer is "not great things": incentive and total pay are inefficiently distorted upwards for the most skilled CEOs, which also comes with significant inefficient manipulation. The general idea goes against conventional wisdom: more competition does not (necessarily) increase economic efficiency.
now forthcoming in Journal of Accounting and Economics.