Aalok Mehta (PhD candidate at USC ) and I just finished a draft of a paper we are scheduled to present at the Telecom Policy Research Conference (TPRC) next month. The paper, “Overestimating Wireless Demand: Policy and Investment Implications of Upward Bias in Mobile Data Forecasts,” is about forecast error in demand estimates for spectrum.
The abstract is below, and a full copy is available at – http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2418364
In this paper, we present evidence of persistent errors in projections of wireless demand and examine the implications for wireless policy and investment. Mobile demand projections are relied upon in academic and government research and used for critically important telecommunications policy decisions, both domestically and internationally. The Federal Communications Commission, for example, used such projections to estimate a 275 MHz spectrum shortage by 2014 and featured such estimates in the U.S. National Broadband Plan as evidence for allocating additional spectrum for cellular services. The International Telecommunications Union Radiocommunication Sector endorsed in 2006 an estimate of a 1,280- to 1,720-MHz spectrum shortage by 2020. In hindsight, both estimates clearly overstated demand; however, they remain the basis for policy direction, and their underlying accuracy has not been evaluated in a systematic manner. The potential for persistent bias in these projections may allow policy errors based on these data to compound over time as opposed to self-correct. Biased industry projections also cause problems for private investment. Investors rely on industry projections as a significant input in their investment decision-making process. An upward bias in wireless projections can result in overallocation of capital in licensed wireless systems versus other industries where, on a marginal basis, capital might be more productive. It may also cause investors to overpay for wireless assets. As this overallocation becomes apparent, prices typically drop and the industry must downsize or correct. This creates unnecessary volatility and may drive away future investments.
To investigate upward biases in wireless projections, we examine major historical projections of wireless demand—including those from Cisco, Coda Research, Yankee Group, and the ITU—and compare them to updated projections over time and actual results. Assuming such projections are neutral, the number of projections that overestimate demand should be roughly equal to those that underestimate demand. Likewise, we would expect that, as projections are updated over time, errors on the upward side should roughly equal those on the downward side. However, the evidence reveals a persistent tendency to overestimate in both number and value. Of the past seven Cisco mobile traffic forecasts for North America, for example, overestimates were nearly twice as frequent as underestimates (19 vs. 10). Overestimates are also on average of greater magnitude than underestimates (103 vs. 81 PB/month). Moreover, the longer the timeframe, the more likely the estimates are likely to be over-projections due to exponential growth issues. Many of these errors result from failures to anticipate or correct accurately for new business practices, such as mobile offloading or tiered data plans.
Wireless policy decisions have long timeframes; more than half of recent U.S. spectrum reallocations have taken a decade or longer, and incumbent users often possess significant technical and political power to stall or limit spectrum reorganizations. Persistent upward biases in mobile demand projections thus have significant, long-term policy implications, including possibly overweighting spectrum policy towards expensive clearings for new licensed bands over more flexible shared or unlicensed management systems. Such biases may also lead to businesses overvaluing spectrum licenses and to governments overestimating the contributions of exclusive spectrum allocations to deficit reduction and economic growth.
Our findings suggest the mobile industry contains much higher levels of inherent demand uncertainty than is commonly estimated and that business and governments may not be fully factoring it into their policy decisions. To reduce dependence on uncertain estimates, government officials should consider assigning spectrum allocations with greater flexibility of use. Additionally, we recommend that policymakers use demand projections only when they are transparent, have authorial accountability, and comply with processes to reduce conflicts of interest.