Pith. sign in

REVIEW 1 major objections 5 minor 50 references

This paper proves that equilibrium fixed-point problems can be solved exactly in a lower-dimensional space whenever the original operator factors as S = D∘E and T = E∘D, a structure it calls strong semiconjugacy.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 19:06 UTC pith:AFHNNZZN

load-bearing objection The strong-semiconjugacy theory is clean and the SA debiasing trick is genuinely useful, but the 70,000x speedup claim does not survive a fair baseline and should be fixed before publication. the 1 major comments →

arxiv 2607.17073 v1 pith:AFHNNZZN submitted 2026-07-19 econ.GN math.OCq-fin.EC

Faithful Decoding

Nisha Peng , John Stachurski , Jingni Yang , Ziyue Yang This is my paper
classification econ.GN math.OCq-fin.EC
keywords strong semiconjugacyencoder-decoder factorizationdimensionality reductionfixed-point transferdynamic programmingasset pricingstochastic approximationreal options
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper seeks to establish that many high-dimensional equilibrium and dynamic-programming problems can be reduced to lower-dimensional fixed-point problems with no information loss. The mechanism is a factorization: if the original operator S can be written as S = D∘E through an encoder E and a decoder D, and the companion operator T = E∘D is easier to solve, then the fixed points of S and T correspond one-to-one, and iterating T while decoding once yields the exact fixed point of S. The paper demonstrates this structure in a real option problem, a turn-based zero-sum stochastic game, an asset pricing model with K assets, and an asset pricing model with heterogeneous beliefs. If the claim is right, economists can replace a high-dimensional operator by a much smaller one in computation and still recover exactly the same solution, with reported speedups up to 70,000 times.

Core claim

The paper's central discovery is the notion of strong semiconjugacy and its transfer theorems. Two dynamical systems (V,S) and (W,T) are strongly semiconjugate under E,D when S = D∘E on V and T = E∘D on W. Theorem 3.1 shows that then the sets of fixed points, fix(S) and fix(T), have the same cardinality, with E giving the bijection. Theorem 3.3 shows that if (W,T) is globally stable and D is continuous, then (V,S) is globally stable with unique fixed point v̄ = D w̄, and D T^n w → v̄ for every w. Local stability transfers when both E and D are continuous. The paper applies this identity to build low-dimensional operators in a real option problem, a turn-based zero-sum game, asset pricing wit

What carries the argument

The central object is the strong-semiconjugacy identity S = D∘E on V and T = E∘D on W, where E is the encoder and D is the decoder. This factorization carries the whole argument: it makes the fixed-point sets of the two systems bijective and converts iterates of T into iterates of S, so stability properties transfer from the easier system to the harder one. It is weaker than topological conjugacy—neither E nor D needs to be bijective—so dimension can drop, yet it is stronger than plain semiconjugacy, so no solution information is lost. In the main applications, E integrates out transient shocks so that T acts only on the persistent state; in the stochastic game, E and D are the two Bellman o

Load-bearing premise

The economic applications rest on the assumption that the state space factorizes as X = Z × E with the transient component ξ iid and independent of the persistent process Z, so that the encoder can integrate out ξ and the low-dimensional operator acts only on Z; without this separability, the exact factorization S = D∘E, T = E∘D no longer reduces dimension.

What would settle it

Rerun the real option experiment with the transient shocks drawn from a Markov chain that is correlated with the persistent state rather than iid and independent. If the minimal exact factorization then forces the 'low-dimensional' operator to keep the transient state, or the decoded iterates D T^n w fail to converge to the unique fixed point of S, the paper's claim that real problems admit lossless dimension reduction would be false; the algebraic transfer theorems themselves would remain true.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • In a real option problem with state separated into persistent Z and iid transient ξ, the value function is exactly computable by iterating the low-dimensional operator on Z and decoding once; the reported speedup grows roughly quartically in the transient grid size and reaches about 70,000 times at grid size 9.
  • Global or local stability of the low-dimensional system transfers to the original system, so existence, uniqueness, and convergence of fixed-point iteration for the hard problem can be proved by studying only the easier system.
  • In a turn-based zero-sum stochastic game, the value on the larger state space is the decoded fixed point of a β²-contraction on the smaller state space, reducing memory from n2 to n1 and time whenever many next-states are unreachable.
  • In heterogeneous-beliefs asset pricing, applying stochastic approximation to the transformed operator T instead of the original max-of-expectations operator eliminates the upward max-of-sample-means bias, and the decoded iterates converge to the equilibrium price.
  • Monotone comparative statics transfer: if parameter shifts make encoders and decoders dominate pointwise and remain order-preserving, the corresponding fixed points move monotonically—for instance, a higher discount factor raises the real-option value.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • An extension left implicit: the factorization is most valuable when the model contains a block of iid shocks independent of the persistent state; such nuisance blocks appear in many applied models, and each could be compressed by the same encoder-decoder recipe.
  • The bias-removal mechanism in the heterogeneous-beliefs application suggests a general rule for stochastic approximation: when the fixed-point map contains an outer max, move the max inside the expectation through a decoder before applying stochastic approximation; the paper demonstrates one instance, but the rule should carry to other max-of-expectations models.
  • If the exact factorization is unavailable, an approximate version with explicit error bounds would connect this approach to lossy aggregation methods and let researchers measure how much information is lost when they compress; the paper does not develop this.
  • The monotonicity theorem is stated for value functions, but the same order-preserving argument applies to any equilibrium object that is monotone in parameters, such as price functions in production networks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 5 minor

Summary. The paper introduces a notion of 'strong semiconjugacy' between dynamical systems (V,S) and (W,T) defined by the factorization S = D∘E and T = E∘D. Theorems 3.1–3.3 show that fixed point sets are in bijection and that local/global stability transfers from the low-dimensional system (W,T) to the high-dimensional system (V,S). The framework is applied to a real option problem (where the encoder integrates out iid transient shocks), a turn-based zero-sum stochastic game, an asset pricing model with a stochastic discount factor, and an asset pricing model with heterogeneous beliefs where a transformed stochastic approximation scheme is claimed to be unbiased. Section 5 adds parametric monotonicity results. The abstract claims speed gains up to 70,000 times in the real option example.

Significance. The theoretical framework is elegant and the fixed-point transfer/stability results are clean and correct. The application to stochastic approximation with heterogeneous beliefs is a genuinely useful contribution: encoding removes the 'max-of-sample-means' bias while preserving the exact fixed point. The monotonicity extension is a nice addition. However, the headline computational speedup claim in the real option example is not supported by the evidence as presented, because the baseline implementation of S does not exploit the paper's own factorization S = D∘E. The theoretical core, the SA bias correction, and the game-theoretic complexity analysis are independent of this timing claim, and they remain valuable.

major comments (1)
  1. [Section 2.2 / Table 1] The reported speedup compares iteration on T against a 'direct' implementation of S that re-integrates over all successor states. But Lemma 3.6 proves S = D∘E, and moreover S^n v0 = D T^{n-1} E v0 for any v0, so the iteration counts coincide whenever the initial encode is performed. An optimized S iteration, using the factorization, costs O(|Z||E| + |Z|^2) per step — the same asymptotic cost as one T iteration (which computes D w then E(D w)). Thus the 70,000× figure in Table 1 is an artifact of the naive baseline; with optimized S the time ratio is O(1). The abstract's headline speed claim is therefore unsupported. Section 4.1 gives a fair comparison because there T genuinely avoids a full decode per iteration; the real-option example should be rebenchmarked against an optimized S, or reframed in terms of memory savings and exactness rather than time speedup.
minor comments (5)
  1. [Section 3.5] The claim that encoding into (bZ,T) 'provides a transparent contraction result' is overstated. For constant β, the original operator S is also a contraction on the complete space ℓX, with modulus β: the same argument used for T applies, because ∥S v1 − S v2∥_ℓX ≤ β∥v1 − v2∥_ℓX. The real benefit of the encoding is the smaller state space, not the creation of a contraction where none existed.
  2. [Section 2.2, text after Table 1] The sentence 'Consequently, the speedup grows roughly quartically with m' describes the naive baseline's complexity ratio. If S is implemented via Lemma 3.6, the ratio is O(1); please clarify that the quartic growth is specific to the deliberately naive S implementation.
  3. [Section 4.3 / Appendix C] Figure 3 is referenced in Section 4.3 but appears only in Appendix C. The placement or numbering should be adjusted so that the reader can find the bias illustration without jumping to the appendix.
  4. [Section 4.3] The sentence 'direct Monte Carlo fails to offer the bootstrapping associated with stochastic approximation methods and hence has relatively high variance' is vague; it would be helpful to state the precise variance comparison or cite a specific result.
  5. [Appendix B] Equation (24) contains unspecified constants C1,C2 and a bound with log(1/δ) and entropy integrals; the dependence on δ and M is clear, but a reference to the exact theorem in Mou et al. (2022) that yields this bound would help the reader verify it.

Circularity Check

0 steps flagged

No significant circularity: fixed-point transfer follows from the semiconjugacy identities and is verified by direct calculation; cited contraction/stability results are external theorems.

full rationale

The derivation chain is self-contained. Theorem 3.1 proves fixed-point transfer directly from the defining identities S = D∘E and T = E∘D, and the proof is a genuine argument rather than a restatement: it shows E is injective on fix(S), so the cardinality conclusion is non-tautological. The applications (Lemmas 3.6, 4.2, 4.4) verify the identities by direct calculation from the economic primitives; they do not fit parameters to the target fixed points. The contraction and global-stability premises used to feed Theorem 3.3 come from Banach's fixed point theorem, Stachurski and Zhang (2021), Borovička and Stachurski (2021), and Mou et al. (2022). Although the first two are co-authored by a co-author of the present paper, they are external published theorems with stated assumptions and are not fit-to-target results, so under the scoring rules they do not raise the circularity score. The only point of concern is the Table 1 timing comparison, which compares a deliberately direct implementation of S against T even though Lemma 3.6 gives S = D∘E; an optimized S using the paper's own factorization would have comparable per-iteration cost. That is an evaluation/benchmark concern about the reported 70,000x speedup, not a circularity in the theoretical derivation, so it is not scored here as a circular step.

Axiom & Free-Parameter Ledger

5 free parameters · 8 axioms · 0 invented entities

The theoretical core rests on standard contraction/spectral results and the domain-specific factorization/independence assumptions; the numerical illustrations introduce hand-chosen calibration parameters but no fitted constants. No invented entities.

free parameters (5)
  • Numerical grid sizes (n_z, m_ξ) = n_z=(4,4,4,4), m_ξ=(m,m,m,m), m=1..9
    Chosen for the speedup experiment; the complexity gain claim depends on m.
  • AR(1) and shock distribution parameters (ρ, ν, μ, σ) = ρ=(0.9,0.9,0.9,0.9), ν=(0.1,0.1,0.2,0.2), μ=0, σ=0.1
    Hand-chosen for the real-option illustration; not fit to empirical data.
  • Economic/policy parameters (θ, β_min, β_max) = θ=0.5, β_min=0.05, β_max=0.95
    Chosen for the illustration; β is rescaled to lie in this range.
  • SA hyperparameters (a0, B0, M) = a0∈{0.8,0.9,1.0}; B0 and M not reported precisely
    Selected to satisfy Mou et al. admissibility; not fixed in the paper, hampering replication.
  • Heterogeneous-beliefs AR parameters = ρ1=ρ2=0.1, ν1=0.015, ν2=0.01
    Chosen for the Figure 5 example.
axioms (8)
  • standard math Banach fixed point theorem (and its use on complete sup-norm spaces)
    Used to establish global stability of T in Sections 3.5, 4.1, 4.3.
  • standard math Theorem 2.1 of Stachurski and Zhang (2021): eventual contraction criterion for state-dependent discounting
    Invoked in Lemma 3.7 to show (bZ,T) globally stable under Assumption 3.1(iii).
  • standard math Theorem C.7 of Borovička and Stachurski (2021): spectral radius ρ(H)<1 under Assumption 4.1
    Invoked in Lemma 4.3 to show T globally stable in the asset pricing application.
  • standard math Theorem 1 of Mou et al. (2022): ROOT-SA convergence in Banach spaces
    Used in Appendix B to translate operator-defect bounds into convergence of D w_M to v̄.
  • domain assumption Kernel factorization N((z,ξ), A×B)=Q(z,A)φ(B) with ξ transient and independent of Z
    The core dimension-reduction identity S=D∘E and T=E∘D in Sections 2 and 4.2 holds only with this product structure.
  • domain assumption Assumption 3.1: bounded measurable π,c, bounded β, eventual contraction condition for T
    Required for well-posedness and global stability in the real option application.
  • domain assumption Assumption 4.1: integrable dividends, irreducible Q with stationary ψ, eventual compactness of H, negative SDF Lyapunov exponent
    Required for global stability of T in the K-asset pricing application.
  • domain assumption Closed partial order and order-preserving E_θ,D_θ in Theorem 5.1
    Needed to transfer monotone parameter shifts from T-fixed points to S-fixed points.

pith-pipeline@v1.3.0-alltime-deepseek · 26841 in / 21037 out tokens · 192710 ms · 2026-08-01T19:06:27.200479+00:00 · methodology

0 comments
read the original abstract

This paper studies transformations that increase efficiency in solving equilibrium systems without information loss. Our approach exploits order-theoretic structure commonly found in economic problems to obtain conditions under which high-dimensional systems can be transformed into low-dimensional systems while preserving exact relationships between their solutions. The transformations can also be used for purposes other than dimensionality reduction, such as simplifying analysis and facilitating stochastic approximation routines. The theoretical ideas are illustrated using applications from economics and finance. In a real option problem, we demonstrate speed gains of up to 70,000 times.

Figures

Figures reproduced from arXiv: 2607.17073 by Jingni Yang, John Stachurski, Nisha Peng, Ziyue Yang.

Figure 1
Figure 1. Figure 1: Left: Speed gain obtained by iterating with 𝑇 instead of 𝑆. Right: Raw computation times for the fixed points of 𝑆 and 𝑇 as a function of the common transient grid size 𝑚𝜉 = (𝑚, 𝑚, 𝑚, 𝑚). In all experiments we compare direct iteration on 𝑆 with iteration on 𝑇 using the same stopping tolerance 𝜀 = 10−6 . Starting from fixed initial iterates 𝑣0 and 𝑤0, for the high-dimensional problem we generate 𝑣𝑞+1 = 𝑆𝑣𝑞 … view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of conjugacy and strong semiconjugacy • deduce the existence of a unique fixed point of (𝑉, 𝑆) only by studying (𝑊, 𝑇), and • compute this unique fixed point, working only with the “nice” system (𝑊, 𝑇). The theoretical results presented in the next section will help us execute this and other closely related tasks. 3.3. Theoretical Results. Our first result provides foundations for the approach o… view at source ↗
Figure 3
Figure 3. Figure 3: Evidence of max-of-means bias: ROOT-SA approximation of the fixed point 𝑣¯. Ground truth 𝑣¯ (blue) vs. approximation using 𝑆ˆ (orange dashed). 0 10000 20000 30000 40000 50000 iteration 10 1 10 0 10 1 10 2 10 3 vt v 2 a0 = 0.8 a0 = 0.9 a0 = 1.0 Dwt Dw 2 [PITH_FULL_IMAGE:figures/full_fig_p033_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Convergence paths for 𝑆ˆ (dashed lines, various 𝛼0) and 𝑇ˆ (black line, 𝛼0 = 1.0) [PITH_FULL_IMAGE:figures/full_fig_p033_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of the true equilibrium price function 𝑣¯ cal￾culated by iterating the operator 𝑆 and the ROOT-SA approximation 𝐷𝑤𝑀. Appendix C. Upward Bias In this section, we illustrate the claim that applying an approximation to equation (22) introduces a systematic upward bias. To see this, we apply the same ROOT-SA algorithm, replacing 𝑇ˆ with 𝑆ˆ: (𝑆𝑣ˆ ) (𝑥) = max 𝑖∈𝐼 𝛽 [PITH_FULL_IMAGE:figures/full_fig_p… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 4 linked inside Pith

  1. [1]

    Bayer, C., Born, B., and Luetticke, R. (2024). Shocks, frictions, and inequality in US business cycles. American Economic Review , 114(5):1211--1247

    2024

  2. [2]

    Bertsekas, D. P. (2017). Dynamic Programming and Optimal Control , volume I. Athena Scientific, 4th edition

    2017

  3. [3]

    Borkar, V., Chen, S., Devraj, A., Kontoyiannis, I., and Meyn, S. (2025). The ODE method for asymptotic statistics in stochastic approximation and reinforcement learning. The Annals of Applied Probability , 35(2):936--982

    2025

  4. [4]

    and Stachurski, J

    Borovi c ka, J. and Stachurski, J. (2021). Stability of equilibrium asset pricing models: A necessary and sufficient condition. Journal of Economic Theory , 193:105227

    2021

  5. [5]

    and Scheidegger, S

    Brumm, J. and Scheidegger, S. (2017). Using adaptive sparse grids to solve high-dimensional dynamic models. Econometrica , 85(5):1575--1612

    2017

  6. [6]

    Day, S., Frongillo, R., and Trevi \ n o, R. (2008). Algorithms for rigorous entropy bounds and symbolic dynamics. SIAM Journal on Applied Dynamical Systems , 7(4):1477--1506

    2008

  7. [7]

    de Vries, J. (2014). Topological Dynamical Systems: An Introduction to the Dynamics of Continuous Mappings , volume 59 of De Gruyter Studies in Mathematics . De Gruyter, Berlin/Boston

    2014

  8. [8]

    Devaney, R. L. (2018). An introduction to chaotic dynamical systems . CRC Press, 2nd edition

    2018

  9. [9]

    Fabianek, P., Glensk, B., and Madlener, R. (2024). A sequential real options analysis for renewable power-to-hydrogen plants for Germany and California . Renewable and Sustainable Energy Reviews , 192:114159

    2024

  10. [10]

    D., Schaal, E., and Taschereau-Dumouchel, M

    Fajgelbaum, P. D., Schaal, E., and Taschereau-Dumouchel, M. (2017). Uncertainty traps. The Quarterly Journal of Economics , 132(4):1641--1692

    2017

  11. [11]

    Farmer, L. E. and Toda, A. A. (2017). Discretizing nonlinear, non- Gaussian Markov processes with exact conditional moments. Quantitative Economics , 8(2):651--683

    2017

  12. [12]

    Forneron, J.-J. (2024). Estimation and inference by stochastic optimization. Journal of Econometrics , 238(2):105638

    2024

  13. [13]

    and Haro, \`A

    Gomis-Porqueras, P. and Haro, \`A . (2009). A geometric description of a macroeconomic model with a center manifold. Journal of Economic Dynamics and Control , 33(6):1217--1235

    2009

  14. [14]

    Gu, Z., Lauri \`e re, M., Merkel, S., and Payne, J. (2024). Global solutions to master equations for continuous time heterogeneous agent macroeconomic models. arXiv preprint arXiv:2406.13726

    Pith/arXiv arXiv 2024

  15. [15]

    Hansen, L. P. and Renault, E. (2010). Pricing kernels. In Cont, R., editor, Encyclopedia of Quantitative Finance . Wiley

    2010

  16. [16]

    Harrison, J. M. and Kreps, D. M. (1978). Speculative investor behavior in a stock market with heterogeneous expectations. The Quarterly Journal of Economics , 92(2):323--336

    1978

  17. [17]

    Hotz, V. J. and Miller, R. A. (1993). Conditional choice probabilities and the estimation of dynamic models. The Review of Economic Studies , 60(3):497--529

    1993

  18. [18]

    constrained optimization approaches to estimation of structural models

    Iskhakov, F., Lee, J., Rust, J., Schjerning, B., and Seo, K. (2016). Comment on “constrained optimization approaches to estimation of structural models”. Econometrica , 84(1):365--370

    2016

  19. [19]

    Kreps, D. M. (1981). Arbitrage and equilibrium in economies with infinitely many commodities. Journal of Mathematical Economics , 8(1):15--35

    1981

  20. [20]

    and Smith, Jr, A

    Krusell, P. and Smith, Jr, A. A. (1998). Income and wealth heterogeneity in the macroeconomy. Journal of Political Economy , 106(5):867--896

    1998

  21. [21]

    and Rugman, A

    Li, J. and Rugman, A. M. (2007). Real options and the theory of foreign direct investment. International Business Review , 16(6):687--712

    2007

  22. [22]

    and Bertsekas, D

    Li, Y. and Bertsekas, D. (2025). An error bound for aggregation in approximate dynamic programming. arXiv preprint arXiv:2507.01324

    Pith/arXiv arXiv 2025

  23. [23]

    and Stachurski, J

    Ma, Q. and Stachurski, J. (2019). Optimal timing of decisions: A general theory based on continuation values. Journal of Economic Dynamics and Control , 101:62--81

    2019

  24. [24]

    and Stachurski, J

    Ma, Q. and Stachurski, J. (2021). Dynamic programming deconstructed: Transformations of the B ellman equation and computational efficiency. Operations Research , 69(5):1591--1607

    2021

  25. [25]

    V., Whittle, A

    Martello, M. V., Whittle, A. J., Oddo, P. C., and de Neufville, R. (2024). Real options analysis for valuation of climate adaptation pathways with application to transit infrastructure. Risk Analysis , 44(5):1046--1066

    2024

  26. [26]

    Mertens, J.-F., Sorin, S., and Zamir, S. (2015). Repeated Games , volume 55 of Econometric Society Monographs . Cambridge University Press

    2015

  27. [27]

    and Shannon, C

    Milgrom, P. and Shannon, C. (1994). Monotone comparative statics. Econometrica , 62(1):157--180

    1994

  28. [28]

    A., Veness, J., Bellemare, M

    Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., et al. (2015). Human-level control through deep reinforcement learning. Nature , 518(7540):529--533

    2015

  29. [29]

    Morris, S. (1996). Speculative investor behavior and learning. The Quarterly Journal of Economics , 111(4):1111--1133

    1996

  30. [30]

    J., Bartlett, P

    Mou, W., Khamaru, K., Wainwright, M. J., Bartlett, P. L., and Jordan, M. I. (2022). Optimal variance-reduced stochastic approximation in Banach spaces. arXiv preprint arXiv:2201.08518

    Pith/arXiv arXiv 2022

  31. [31]

    Munkres, J. (2000). Topology . Prentice Hall, 2nd edition

    2000

  32. [32]

    and Scheinkman, J

    Nutz, M. and Scheinkman, J. A. (2020). Shorting in speculative markets. The Journal of Finance , 75(2):995--1036

    2020

  33. [33]

    and Robinson, B

    Ocampo, S. and Robinson, B. (2024). Computing longitudinal moments for heterogeneous agent models. Computational Economics , 64(3):1891--1912

    2024

  34. [34]

    Payne, J., Rebei, A., and Yang, Y. (2025). Deep learning for search and matching models. Swiss Finance Institute Research Paper 25-05, Swiss Finance Institute

    2025

  35. [35]

    Pohl, W., Schmedders, K., and Wilms, O. (2018). Higher order effects in asset pricing models with long-run risks. The Journal of Finance , 73(3):1061--1111

    2018

  36. [36]

    and Wierman, A

    Qu, G. and Wierman, A. (2020). Finite-time analysis of asynchronous stochastic approximation and Q -learning. In Proceedings of the Thirty Third Conference on Learning Theory , volume 125 of Proceedings of Machine Learning Research , pages 3185--3205. PMLR

    2020

  37. [37]

    T., Fonoberova, M., Mohr, R., Kevrekidis, I

    Redman, W. T., Fonoberova, M., Mohr, R., Kevrekidis, I. G., and Mezi \'c , I. (2022). Algorithmic (semi-) conjugacy via Koopman operator theory. In 2022 IEEE 61st Conference on Decision and Control (CDC) , pages 6006--6011. IEEE

    2022

  38. [38]

    Rust, J. (1987). Optimal replacement of GMC bus engines: An empirical model of Harold Zurcher . Econometrica , 55(5):999--1033

    1987

  39. [39]

    Scheinkman, J. A. and Xiong, W. (2003). Overconfidence and speculative bubbles. Journal of Political Economy , 111(6):1183--1220

    2003

  40. [40]

    Shapley, L. S. (1953). Stochastic games. Proceedings of the National Academy of Sciences , 39(10):1095--1100

    1953

  41. [41]

    and Vieille, N

    Solan, E. and Vieille, N. (2015). Stochastic games. Proceedings of the National Academy of Sciences , 112(45):13743--13746

    2015

  42. [42]

    and Zhang, J

    Stachurski, J. and Zhang, J. (2021). Dynamic programming with state-dependent discounting. Journal of Economic Theory , 192:105190

    2021

  43. [43]

    and Lawrence, J

    Stroombergen, A. and Lawrence, J. (2022). A novel illustration of real options analysis to address the problem of probabilities under deep uncertainty and changing climate risk. Climate Risk Management , 38:100458

    2022

  44. [44]

    Szepesv \'a ri, C. (2010). Algorithms for Reinforcement Learning . Synthesis Lectures on Artificial Intelligence and Machine Learning. Morgan & Claypool Publishers

    2010

  45. [45]

    Tauchen, G. (1986). Finite state Markov -chain approximations to univariate and vector autoregressions. Economics Letters , 20(2):177--181

    1986

  46. [46]

    Topkis, D. M. (1998). Supermodularity and Complementarity . Princeton University Press, Princeton, NJ

    1998

  47. [47]

    van Hasselt, H., Guez, A., and Silver, D. (2016). Deep reinforcement learning with Double Q -learning. In Proceedings of the AAAI conference on artificial intelligence , volume 30, pages 2094--2100

    2016

  48. [48]

    van Hasselt, H. V. (2010). Double Q -learning. In Advances in Neural Information Processing Systems , volume 23, pages 2613--2621

    2010

  49. [49]

    Wibault, C., Forkel, J., Towers, S., Wibault, T., Duque, J., Whittle, G., Schaab, A., Yang, Y., Wang, C., Osborne, M., Moll, B., and Foerster, J. (2026). Recurrent structural policy gradient for partially observable mean field games. arXiv preprint arXiv:2602.20141

    Pith/arXiv arXiv 2026

  50. [50]

    Yang, Y., Wang, C., Schaab, A., and Moll, B. (2025). Structural reinforcement learning for heterogeneous agent macroeconomics. arXiv preprint arXiv:2512.18892

    arXiv 2025