How do cloud computing providers match volatile demand with fixed capacity?

To match volatile demand with fixed capacity, cloud computing platforms employ tiered reliability—offering discounted spot compute services from which users can be “evicted” (i.e., interrupted) with little warning when capacity tightens. We study this market design using proprietary data from a major cloud platform, exploiting a price experiment and the quasi-random nature of evictions to estimate a structural model. The price elasticity of demand is -0.5, and evictions persistently reduce usage by 40%, indicating a strong revealed preference for reliability. More usage increases eviction rates, consistent with congestion. We interpret these facts through a model where heterogeneous users choose the compute reliability for each workload, while learning about eviction risk through experience. On the supply side, evictions arise endogenously given fixed capacity. Preliminary counterfactual results suggest that tiered reliability provides Pareto gains relative to simply allowing the market to clear through congestion.

How do productivity gains from successive generations of AI coding tools translate into shipped software?

How do the productivity effects of AI evolve across successive generations of tools, and to what extent do task-level gains ultimately translate into final output? We study these questions in the context of software development, using data on more than 500,000 GitHub developers combined with their AI usage telemetry. In a matched event study design, we find that autocomplete, interactive coding agents, and autonomous coding agents each significantly increase coding activity (“commits”), with respective cumulative effects of 30%, 180%, and 240%. These gains, however, attenuate sharply across the production hierarchy: the 240% cumulative effect falls to 80% for the number of projects, and to 30% for actual releases. This pattern is consistent with the weak-link hypothesis: the strong productivity gains from AI are attenuated by human bottlenecks in the production chain, with an estimated elasticity of substitution of 0.23 between AI and human effort, which indicates strong complementarities. We further confirm these results across four major software marketplaces, finding a sharp increase in the number of new apps but no increase in total usage. Large task-level AI productivity gains have therefore translated only partially into shipped and used software thus far.

Which economic channels are responsible for Google's high market share in web search?

We study the forces behind Google’s large web-search market share. We develop a demand model with switching costs, quality beliefs, and inattention, and estimate it using a field experiment. We find that (i) requiring active choice barely increases Bing’s market share; (ii) Google users paid to try Bing update positively about its quality and many prefer to continue using it; (iii) many Google users defaulted into Bing do not switch back, consistent with inattention. Counterfactuals suggest that eliminating demand frictions doubles Bing’s market share. Successful remedies expose users to alternative search engines, while data sharing mandates have small effects.

Do recommendation algorithms give digital platforms an ability to self-preference and shape consumer search?

This paper investigates concerns that vertically integrated platforms like Amazon steer demand towards their own offers via algorithmic prominence, potentially harming consumers. On Amazon, for each product, the Buybox prominence algorithm selects one seller to feature, influencing which offers consumers consider. Using novel Amazon sales and Buybox (prominence) data, we estimate a structural model capturing the effects of such algorithmic prominence on consumer choices, seller pricing, and entry. We find that the platform can indeed steer demand as 95% of consumers consider only the Buybox offer. The Buybox is highly price-elastic (−21), but skews towards Amazon’s own offers, which are featured as frequently as observably similar offers priced 5% cheaper. Still, as consumers prefer these offers, this skew does not amount to self-preferencing in the sense of harming consumers: consumer surplus is roughly maximized at the estimated Amazon Buybox advantage, which balances higher prices against showing consumers their preferred offers.

Ukrainean e-procurement auction design allows detection of colluders via re-bidding behaviour.

We develop a method for detecting cartels in multistage auctions. Our approach allows a firm to be collusive when facing members of its cartel yet competitive when facing others. Intuitively, as initial bids are shaded, close initial bids not only imply similar costs but also provide an incentive to undercut. We detect firm pairs that ignore this incentive when facing each other. Our algorithm predicts Ukraine's Antimonopoly Committee's sanctions: firm pairs classified as collusive are 8.98 times more likely (standard error 2.65 times) to be sanctioned. It also uncovers additional collusion: 1,857 collusive firms participate in 15.57% of auctions, increasing costs by 1.95%.

Does algorithmic pricing allow e-commerce sellers to raise prices above the competitive level?

As the economy digitizes, menu costs fall, and firms can more easily monitor prices. These trends have led to the rise of automated pricing (and re-pricing) tools. We employ a novel e-commerce dataset to examine the effect of algorithmic pricing in the wild. Evidence from an event study suggests that firms that start employing repricing tools drop their prices by 16.93%, with market prices falling by 9.67%. However, algorithmic pricing companies have developed ‘resetting’ strategies (which regularly raise prices in the hope that competitors will follow) in order to avoid stark Bertrand-Nash competition. We find that these strategies are effective at coaxing competitors to raise their prices: when a resetting strategy is adopted on a market with less than six serious competitors, both competitor prices and market prices eventually increase by 11.4%. While the resulting patterns of cycling prices are reminiscent of Maskin-Tirole’s Edgeworth cycles, a model of equilibrium in delegated strategies fits the data better. This model suggests that the average price over the cycle will be the monopoly price. Moreover, if the available repricing technologies remain fixed, cycling and prices could rise significantly. However, cycling is still relative rare in the data, even when studying a convenience sample of products with at least one merchant using a repricing tool.

Noisy field-experimental estimates suggest AI-based coding assistants raise developer productivity by 26%.

This study evaluates the effect of generative AI on software developer productivity via randomized controlled trials at Microsoft, Accenture, and an anonymous Fortune 100 company. These field experiments, run by the companies as part of their ordinary course of business, provided a random subset of developers with access to an AI-based coding assistant suggesting intelligent code completions. Though each experiment is noisy and results vary across experiments, when data is combined across three experiments and 4,867 developers, our analysis reveals a 26.08% increase (SE: 10.3%) in completed tasks among developers using the AI tool. Notably, less experienced developers had higher adoption rates and greater productivity gains.

Voter's uncertainness about each other's preferences can be exploited to cheaply buy votes.

We analyze a vote-buying setup where a committee votes on a proposal important to the vote buyer. We characterize the cheapest combination of bribes that guarantees the proposal's passing in different voting environments. We find that the vote buyer publicly offers small bribes to a large supermajority of members for both simultaneous and sequential votes. Each member accepts because he anticipates that the proposal will pass regardless of his vote. We discuss the committee design that maximizes capture cost: combining demanding majority requirements with diversity among members makes the committee more expensive. In small committees, sequential voting increases cost, but the opposite is true for large committees. On the other hand, additional members and transparent voting rules lower the cost. (PDF of Old Version with Additional Examples.)