THE BOUNDED AGENT · THE CANON NESTED
Foundations: Equilibrium as a Limit
The programme's first principle. The classical competitive canon, Walrasian demand, Nash equilibrium, general equilibrium and geometric discounting, is recovered exactly as the infinite-capacity limit of a finite-information agent. These manuscripts lay the microfoundation the rest of the corpus stands on, so they open the treatise.
WORKING PAPER19 pp
Avishek Bhandari
What is a market equilibrium, the state where every plan agrees and nothing moves? This paper argues it is the still shadow of a restless, adaptive economy. By placing the whole economic state, agents, firms, machines, and prices, in a single geometric space where closeness means statistical correlation, it shows that competitive equilibrium is the projection of the economy onto its unchanging part, and the living economy is what is left over. One geometric step recovers three classical ideas at once: equilibrium, rational expectations, and the welfare benchmark. The leftover part measures an economy's ignorance and the information an agent must pay to describe it. The work is theory, with a worked example for illustration and no empirical test, careful that a unique projection is not a determinate equilibrium.
WORKING PAPER22 pp
Avishek Bhandari
Why do people value the present over the future? Economics assumes this impatience without explaining it. This paper derives it from one idea: people process information at a limited rate. To plan for a future payoff you must keep pace with the novelty in your income, and spreading limited attention across the horizons ahead produces a steady discount on the future, one that steepens with both how much novelty the income generates and how costly attention is. Impatience is thus the price of a novel future met by a finite mind, and it disappears when the world is predictable or attention is unlimited. The same idea recasts present bias, the pull toward the near term, as a novelty rate not yet learned, and explains why consumption reacts only gradually and smoothly to income surprises. The account concerns what generates discounting and what the data can pin down, not forecasting, and rests on a reproducible numerical illustration, not an empirical test.
WORKING PAPER16 pp
Avishek Bhandari
Any agent that learns to act faces a bandwidth limit: it can only carry so much about the world into each decision, so its behaviour is a lossy summary of what it sees. This paper treats that information budget as a scarce resource and shows what optimal learning becomes once it is priced. Four familiar objects, the optimal policy, the state representation, the learning dynamics, and the generalisation gap, turn out to be one family of rate-versus-fidelity trade-offs, each collapsing to the textbook version when information is free. A small, fully reproducible tabular study illustrates each result. The results are shown for finite tabular problems and Gaussian representations, and one limit is firm: more capacity buys sharper control, never foresight of a genuinely random world.
arXiv · openecon.TH
Avishek Bhandari
The competitive equilibrium of general equilibrium theory exists as a fixed point and is, by the theory's own results on aggregate excess demand, in general silent on whether that fixed point is unique, stable, or attained. This paper takes the economy to be not a configuration to be solved for but a process to be recovered, an asymptotically mean stationary information source carrying a partially identified operator of statistical dependence, populated by agents that are finite-capacity information channels. Within this adaptive order the competitive, rational expectations equilibrium is recovered exactly, as a joint limit taken along an explicit scaling path. Three parameter limits and two fixed-point conditions deliver it, the entropy rate falls to zero, agent channel capacity diverges, selection intensity grows infinitely sharp, adaptive learning reaches its expectationally stable rest point, and the recovered structure ceases to coevolve. At that corner the limiting object satisfies the axioms of the canon and its rest state is a Walrasian equilibrium, away from it the adaptive economy is a strict generalisation, carrying a positive entropy rate and a recovered dependence structure that the equilibrium primitive cannot express. We give the nesting as a theorem, establish the result by result correspondence with existence, with the Sonnenschein Mantel Debreu indeterminacy, and with the regular economies recovery, and characterise exactly what the equilibrium limit erases.
WORKING PAPER27 pp
Avishek Bhandari
Standard economics assumes the shopper is a flawless calculator who always buys the best basket it can afford. This paper models the shopper instead as a limited information channel: it compresses its world to the detail its attention affords, so its choice is a probability distribution, not a single basket. The textbook consumer returns exactly as the unlimited-attention limit, while at the zero-attention end the shopper falls back on pure habit. The central result is about how this shopper's demand responds to price changes. That pattern of responses is just a rescaling of how the shopper's own choices vary and move together, so it comes out symmetric. And provided the budget really binds, because the shopper wants more than it can afford, raising a good's own price lowers demand for it once buying power is held fixed. So the downward pull comes from the budget and from compression, not from rationality. The framework also covers an artificial agent running a limited-capacity policy. A two-good case is worked out fully in closed form; the paper is theoretical and offers no empirical test.
WORKING PAPER25 pp
Avishek Bhandari
Standard game theory assumes players optimise perfectly. This paper rebuilds it on a player of finite information capacity, whose best response softly favours better actions. That response is the familiar logit rule, so equilibrium becomes the logit quantal response equilibrium, and Nash equilibrium reappears in the unlimited-capacity limit; potential games add a principled equilibrium selection valid only within that class. The gain is a discipline for network games. When the interaction network is inferred from how players co-move rather than assumed, only its leading eigenvalue and, for complements, its centrality ranking are firmly pinned. The leading direction can be targeted; the individual link and the key player cannot, and are refused. The results are theoretical, shown by a closed-form pair of networks that look identical yet reverse every link.
