The Hutchins Roundup brings the latest thinking in fiscal and monetary policy to your inbox. Have something you'd like us to include in the next Roundup? Email us and we'll take a look.
This edition was written by Adriana Adames Acosta, Chesapeake Dowdy, Louise Sheiner, and Jack Spira
Firms racing to secure dominant positions in artificial intelligence may invest more than is socially efficient, leaving the sector vulnerable if the technology’s boost to productivity falls short of expectations. Using a dynamic model calibrated to company balance sheets and disclosed deals, PhurichaiRungcharoenkitkul of the Bank for International Settlements finds that competition encourages firms to commit capital early to gain an advantage over rivals. In the model’s baseline calibration, investment reaches about 1.5 times the efficient level—and roughly three times that level when demand for AI services is less responsive to prices. The author also documents the use of debt and circular financing arrangements in the current AI boom, including deals in which computing providers invest in AI labs that commit to purchase their services. Incorporating these arrangements into the model, Rungcharoenkitkul finds that they can accelerate the buildout but also transmit financial stress across firms if the boom turns to bust. Because AI hardware is specialized, a downturn could force firms to sell assets at fire-sale prices, magnifying losses among indebted companies. The model suggests that the larger and more interconnected the investment boom becomes, the greater the productivity gains required to sustain it—and the more disruptive a reversal could be.
The Liquidity Coverage Ratio (LCR), introduced following the Global Financial Crisis, requires large banks to hold enough high-quality liquid assets to cover projected net cash outflows under a 30-day stress scenario. Banks may hold liquidity above the regulatory minimum, known as an LCR buffer. R. Matthew Darst and co-authors at the Federal Reserve Board examine how banks’ liquidity affected lending during the acute phase of the COVID-19 crisis, when corporate borrowers engaged in a “dash for cash” by drawing on pre-established credit lines with banks. They find that banks with above-median LCR buffers provided 10.5% more credit than banks with smaller buffers to firms with substantial unused credit lines—the borrowers with the greatest scope to draw additional credit. Furthermore, only buffers, not overall LCR levels, predicted lending. The authors argue that, although the LCR requirement increased the level of liquidity on bank balance sheets, banks treated the liquidity needed to satisfy the requirement as unavailable for additional lending.
In the 1970s, political divisions over the Vietnam War led Congress to sharply reduce the U.S. military's funding of scientific research. Combining data on Department of Defense research contracts; university finances, employment, and PhD production; individual scientists’ funding sources and employment histories; and records of national publications and patents, Daniel P. Gross and Hansen Zhang of Duke and Bhaven N. Sampat of Johns Hopkins find that fields with more military funding prior to 1970 contracted significantly relative to those with less. In these fields, university departments shrank and produced fewer PhDs, early-career scientists left academia for industry, and national publishing and patents declined relative to other countries. Looking at outcomes more closely connected to the military’s mission, the authors find that the funding cuts led to fewer PhDs entering defense contracting over time and, after an initial increase when scientists moved into industry, a long-run decline in defense-sector patents relative to other technologies. They warn that "scientific capacity might be easier to dismantle than build: declines in public investment may alter the size, composition, and impacts of the scientific sector in ways that persist long after research policy stabilizes."
"A related consideration is the potential effect of AI on the longer-run neutral rate of interest. Often called r*, this is the real interest rate consistent with the economy operating at its full potential once all shocks have dissipated. If AI leads to permanently higher levels of productivity growth, it may increase firms' desire to invest and, hence, their demand for funding. Higher productivity growth may also discourage household savings by increasing expected future income. Under these circumstances, to reconcile the increase in investment with reduced savings, r* would likely rise," says Philip Jefferson, vice chair of the Federal Reserve.
"However, predicting changes in the neutral rate is challenging, given the historically noisy relationship between productivity growth and real interest rates. Furthermore, potential AI-induced increases in inequality could have mitigating effects on r*. High-income households tend to save at higher rates than low-income households. Thus, a rise in income inequality could lead to an increase in the supply of savings, putting downward pressure on the neutral rate.
"While r*, like potential output, is not directly observable, estimating it is of considerable importance for monetary policymakers, because these estimates are informative about the range of interest rates that we consider broadly neutral. If AI indeed raises the neutral rate, then for any given level of the federal funds rate, policy effectively becomes more accommodative. Conversely, if AI-related developments cause r* to decline, policy effectively becomes more restrictive."
About the Hutchins Center on Fiscal and Monetary Policy at Brookings
