Search

Two issues of algorithmic collusion are addressed in this paper. First, we show that in a general class of symmetric games, including Prisoner's Dilemma, Bertrand competition, and any (nonlinear) mixture of first and second price auction, only (strict) Nash Equilibrium (NE) is stochastically stable. Therefore, the tacit collusion is driven by failure to learn NE due to insufficient learning, instead of learning some strategies to sustain collusive outcomes. Second, we study how algorithms adapt to collusion in real simulations with insufficient learning. Extensive explorations in early stages and discount factors inflates the Q-value, which interrupts the sequential and alternative price undercut and leads to bilateral rebound. The process is iterated, making the price curves like Edgeworth cycles. When both exploration rate and Q-value decrease, algorithms may bilaterally rebound to relatively high common price level by coincidence, and then get stuck. Finally, we accommodate our reasoning to simulation outcomes in the literature, including optimistic initialization, market design and algorithm design.

As firms' pricing strategies increasingly rely on algorithms, two concerns have received much attention: algorithmic tacit collusion and price discrimination. This paper investigates the interaction between these two issues through simulations. In each period, a new buyer arrives with independently and identically distributed willingness to pay (WTP), and each firm, observing private signals about WTP, adopts Q-learning algorithms to set prices. We document two novel mechanisms that lead to collusive outcomes. Under asymmetric information, the algorithm with information advantage adopts a Bait-and-Restrained-Exploit strategy, surrendering profits on some signals by setting higher prices, while exploiting limited profits on the remaining signals by setting much lower prices. Under a symmetric information structure, competition on some signals facilitates convergence to supra-competitive prices on the remaining signals. Algorithms tend to collude more on signals with higher expected WTP. Both uncertainty and the lack of correlated signals exacerbate the degree of collusion, thereby reducing both consumer surplus and social welfare. A key implication is that the over-usage of data, both payoff-relevant and non-relevant, by AIs in competitive contexts will reduce the degree of collusion and consequently lead to a decline in industry profits.

We use a geometric model for the hadron polarization with an emphasis on the rapidity dependence. It is based on the model of Brodsky, Gunion, and Kuhn and that of the Bjorken scaling. We make predictions for the rapidity dependence of the hadron polarization in the collision energy range 7.7-200 GeV by taking a few assumed forms of the parameters. The predictions can be tested by future experiments., Comment: RevTex 4, 16 pages, 11 figures

We propose alternative methods to measure the global polarization of $\Lambda$ hyperons. These methods involve event averages of proton's and $\Lambda$'s momenta in the lab frame. We carry out simulations using these methods and show that all of them work equivalently well in obtaining the global polarization of $\Lambda$ hyperons., Comment: Revtex 4, 7 pages, 2 figures; To be published in Chinese Physics C

We study the dynamics of charm quarks in the partonic medium and its implication to the di-electron spectra in high energy heavy ion collisions. The charm quarks traversing a thermal medium is simulated by the relativistic Langevin equation for elastic scatterings of charm quarks by thermal partons in an expanding fireball. The transport coefficients of charm quarks are calculated by the in-medium T-matrix method, where a static heavy quark potential is used with parameters fitted by the lattice QCD results. The di-electron invariant mass spectra are computed in most central collisions and are compared to the STAR data. The angular correlations of di-electrons are almost the same in $p+p$ and Au+Au collisions in the mass range $1.1