National Bureau of Economic Research: Earnings Manipulation and Managerial Investment Decisions: Evidence From Sponsored Pension Plans [PDF--50 pages, 406 KB]. Excerpts: Manipulation of reported earnings can be a powerful tool for managers to inflate their stock prices. Studies of earnings manipulation typically emphasize aggregate measures such as accruals and, consequently, struggle to link earnings manipulation to investment decisions. In this paper, we identify a simple mechanism for earnings manipulation, describe how manipulation through this channel is linked to CEOs' incentives, and show that firms change investment decisions in order to justify, and capitalize on, this type of earnings manipulation. Specifically, we find that managers opportunistically choose assumed rates of return on pension assets, and we examine how these distorted reporting decisions interact with option exercises, merger activity and asset allocation decisions within pension plans. [...]

Some capital market observers have viewed the actions of IBM, under CEO Louis Gerstner, Jr., as an example of how firms can use pension accounting to manipulate earnings. IBM sponsors a large defined benefit pension plan, with over $57 billion in assets at the end of 2002. Table 1 outlines the operating performance of IBM, the performance of its DB pension plan, and the CEO's option grants and exercises. Changes in the long-term rate of return (LTROR) that IBM assumes on its DB pension plan assets are of particular interest. IBM changed its assumed long-term rate of return four times during this period: a twenty-five basis point reduction in 1995, a twenty-five basis point increase in 1997, a fifty basis point increase in 2000, and a fifty basis point reduction in 2002. As we describe more fully in the sections that follow, IBM's assumed rates of return throughout this period exceeded those used by most firms. The frequent changes are also notable given the long run nature of these assets and assumptions.
While IBM reacted to poor actual performance in its pension plan in the mid-1990s by reducing the assumed long-term rate of return, the opposite occurred in 2000. Despite poor equity market returns and declining bond yields during that year, IBM raised its long-term rate of return assumption by fifty basis points. Nearly five percent of IBM's income before tax in 2000 and 2001 resulted from the increase in the assumed long-term rate of return from 9.25% to 10.00%. More generally, IBM's reported pretax income grew at a compound annual growth rate of 6.7% from 1995 to 2001; without these changes, income would have grown at only a 5.6% rate. As Table 1 shows, these changes in pension assumptions coincided with deteriorating operating performance. This example illustrates how senior managers can use pension accounting to boost their firms' reported profits. [...]
We conclude by framing our investigation of earnings manipulation within the debate on whether earnings manipulation reflects an agency concern or is beneficial to current shareholders. We show that managers who are the least constrained by their shareholders - as measured by an index of corporate governance - appear to be the most aggressive with their rate of return assumptions. This evidence suggests that the earnings manipulation investigated here does not benefit current shareholders. We go on to speculate on the magnitude of these effects by returning to the case of IBM. We estimate that between $12 and $76 million of compensation accrued to Gerstner from these changed assumptions alone.