A new post on Econbrowser, titled "The Marginal Efficiency of Investment Equation Interpreted in 2024-26," examines how the United States has recorded elevated nonresidential fixed investment even as interest rates have moved higher. The author's answer: the investment function itself has shifted upward, driven by improved prospects for profits from artificial intelligence.

The analysis is set in the IS-LM model, which uses an investment function linking business spending to income and the cost of borrowing. The post employs the modification introduced by economist Olivier Blanchard — a former chief economist of the International Monetary Fund whose macroeconomics textbook is widely used — under which investment rises with income, an accelerator-type relationship.

The central puzzle the author identifies is how to reconcile higher nonresidential fixed investment with higher interest rates, aside from the endogeneity of the federal funds rate — the observation that the policy rate often moves up because the economy is already strong. One resolution, the post argues, is to recognize that the constant term b0 in the investment function is not necessarily fixed and can be thought of as changing exogenously. The 2024-26 example cited is the increase in the prospects of AI for future profits, which raises investment demand at any given level of interest rates.

To illustrate the point, the post compares total nonresidential fixed investment with the same series excluding computers and software, both normalized by GDP, against interest rates. The underlying data come from the Bureau of Economic Analysis, the US Treasury, the Federal Reserve, and the National Bureau of Economic Research (NBER).

Figure 1: Nonresidential fixed investment to GDP (black, left scale), nonresidential fixed investment ex-computers and software to GDP (blue, left scale), the 10 year Treasury rate, % (red, right scale), and Fed funds rate, % (pink, right scale). NBER defined recessions shaded gray. Source: BEA, Treasury, Fed, NBER.

The key observation is that, outside of computers and software, nonresidential fixed investment has declined as a share of GDP, meaning that investment in computing equipment and software accounts for the divergence between aggregate investment and prevailing interest rates. This decomposition distinguishes the overall investment picture from the portion most directly associated in the analysis with the recent increase in AI-related spending.

The concept at stake traces back to John Maynard Keynes, who in his 1936 "General Theory of Employment, Interest and Money" defined the marginal efficiency of capital — the expected return on an additional unit of investment — as declining as investment expands. John Hicks formalized Keynes's ideas into the IS-LM framework in 1937, and the model remains a staple of macroeconomic teaching, with investment typically modeled as falling when interest rates rise.

The 2024-26 period provides an unusual test of that inverse relationship. The Federal Reserve raised its policy rate to its highest levels in more than two decades during 2022-23 to combat inflation and began lowering it in September 2024, while businesses have sharply increased capital spending on data centers, semiconductors, and other AI-related infrastructure since the public debut of generative AI tools such as ChatGPT in late 2022.

The full post is available on Econbrowser.