Is price-impact decay statistically identifiable by propagator estimation? Diffusive prices force the inferred kernel to be transient and fine-tuned

arXiv:2611.10574 · q-fin.TR, cond-mat.stat-mech, physics.soc-ph, q-fin.MF, q-fin.ST · Submitted 2026-11-07 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "Is price-impact decay statistically identifiable by propagator estimation? Diffusive prices force the inferred kernel to be transient and fine-tuned".

Warren: Here's the gist of the paper. Is price-impact decay statistically identifiable by propagator estimation?

Ruth: This paper proves that asymptotic impact decay is not statistically identifiable by propagator estimation because for diffusive prices, the fine-tuned decay is returned whether or not the true impact decays The Core Argument The central finding is a no-go theorem demonstrating that asymptotic impact decay cannot be identified from propagator estimation when prices are diffusive because for such processes, the fine-tuned decay is an artifact of the deconvolution rather than evidence of true impact decay This result rests on universal theorems for Wiener–Hopf deconvolution which state that the inferred propagator is necessarily transient, and its decay exponent is fixed by the order-flow autocorrelation alone, irrespective of the true impact The authors demonstrate this using an exactly solvable null model where impact is strictly permanent, yet the inferred propagator shows an artificial fine-tuned decay Universal Decay Theorems The paper establishes universal decay theorems for the deconvolution estimation under diffusive price dynamics and long-memory order flow, stating that the inferred kernel is necessarily transient with an asymptotic form of either G(τ) = AGτ−βc + o(τ−βc) or G(τ) = AGτ−βc ln τ + o(τ−βc ln τ), where βc is fixed by the order-flow autocorrelation alone These theorems apply even to a null model with strictly permanent impact and no decay, such as the Sato–Fujiwara–Kanazawa (SFK) model The theorems show that the property of fine-tuned decay is a property of the deconvolution, not evidence of impact decay Model Class Equivalence The non-identifiability arises because the estimand TP = (G∞, β) depends only on γ within the diffusive class MD, meaning it carries no information beyond γ for any prior on MD The study shows that the linear propagator class and models outside this class, such as the SFK model with strictly permanent impact, are observationally equivalent with respect to TP because both map to inferred kernels with G∞ = 0 and the same decay exponent βc This fragility under model misspecification suggests that the inferred decay has not been subjected to a robustness check against generative models outside the propagator class Nonlinear Estimation and Proxies The non-identifiability persists even when moving to nonlinear (quasi-linear) propagator estimation because it reduces to the same Wiener–Hopf deconvolution after a change of variables Furthermore, using metaorder proxies reconstructed via daily segmentation in the SFK model introduces an artificial long memory, which causes the universal decay theorems to apply and confirm that agreement between an inferred propagator and fine-tuned decay law cannot be taken as evidence for transient impact The analysis confirms that the estimand TP = (G∞, β) identifies none of the underlying mechanisms, nor impact decay itself The conclusion is that agreement between an inferred propagator G(τ) and a fine-tuned decay law cannot be taken as evidence for transient impact This suggests a diagnostic where the inferred exponent is compared with the value fixed by the input autocorrelation alone, and this agreement carries no information about the mechanism The inferred kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions SFK Model Validation The numerical validation using the SFK model confirms that even though it has no true decay, its inferred kernel still shows an artificial fine-tuned decay G(τ) ∝ τ−βc This confirms the prediction of universal decay theorems for the SFK model The analysis of the inferred propagator Gn shows that G∞ = 0, implying that the inferred propagator is transient The coefficient AG is shown to be non-zero, which implies AG = B(βc, γ) sin(πβc)/ (2 AS AC) > 0 Asymptotic Behavior of Gn The asymptotic behavior of the inferred propagator Gn is characterized by limn→∞ n βc ln n Gn = AG, where AG is given by B(βc, γ) sin(πβc)/ (2 AS AC) The analysis splits Gn into three parts: G1(n), G2(n), and G3(n), showing that only the leading term G1(n) converges to AG, while the lower-order terms G2(n) and G3(n) vanish as n → ∞ This confirms that the inferred propagator Gn is transient Final Identification Diagnostic The paper suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone This comparison carries no information about the mechanism, serving as a prompt to test the estimation outside the fitted class The inferred fine-tuned decay is a property of the estimator before it is a finding about the dynamics The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimator TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself The universality theorem is the mirror image of a no-go theorem for identifying microscopic dynamics from the universal observables alone In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself <ref:2611.

Paper summary: Ruth: So, we're looking at this paper called "Is price-impact decay statistically identifiable by propagator estimation? Diffusive prices force the inferred kernel to be transient and fine-tuned". Basically, they're asking if we can tell if there's actual impact decay in markets just by looking at how a certain propagator model estimates the price dynamics.

Warren: That sounds like a deep dive into whether a specific tool can actually find what it's supposed to find, or if it just finds some kind of artifact.

Iris: It claims that for diffusive prices, you can't statistically identify asymptotic impact decay using this method because the fine-tuned decay you see is just an artifact of the deconvolution process itself, not evidence of real impact decay.

Ruth: The core idea seems to be a no-go theorem based on universal theorems from Wiener–Hopf deconvolution. They argue that when prices are diffusive, the inferred propagator is forced to be transient, and its decay rate is locked in by the order flow autocorrelation alone, regardless of whether the true impact actually decays or not.

Warren: That's a pretty strong claim. It suggests that if you see a certain decay pattern in your data from this estimation method, you can't be sure if it’s real market impact or just something the math does to the deconvolution process itself.

Iris: They show this using an exactly solvable null model, specifically the Sato–Fujiwara–Kanazawa or SFK model where impact is strictly permanent—meaning it never decays at all. Even in that case, their inferred propagator shows a fine-tuned decay of G(tau) proportional to tau minus beta c, which they call artificial only in that test setting because the true impact doesn't decay >

Ruth: So, they use this null model to show that even when the underlying mechanism has no decay, the method still spits out a specific kind of transient behavior. This is important because it highlights how fragile our identification methods are when we don't perfectly match our models to reality.

Warren: It means if you see that fine-tuned decay in your estimates, you have to be careful not to mistake that for actual impact decay in the market structure itself because the method itself imposes a transient shape on the result >

Iris: The paper establishes universal decay theorems for this deconvolution estimation under diffusive price dynamics and long memory order flow. These theorems state that the inferred kernel is necessarily transient with an asymptotic form of either G(tau) proportional to tau minus beta c plus some lower order terms, or G(tau) proportional to tau minus beta c times the natural logarithm of tau plus some lower order terms >

Ruth: That's a very specific statement about the universal behavior. They fix that decay exponent, beta c, based only on the order flow autocorrelation alone and not on any external information about market impact decay >

Warren: So, the key finding is that for diffusive prices and long memory, the estimand they are looking at—which involves G infinity and beta—only tells you about gamma within that diffusive class MD, meaning it carries no extra information beyond gamma >

Iris: They demonstrate this by showing that linear propagator models and models outside that class, like the SFK model with strictly permanent impact, are observationally equivalent regarding the estimand TP because both map to inferred kernels with G infinity equal to zero and the same decay exponent beta c >

Ruth: It suggests a kind of robustness check is missing. The fragility they find in this estimation method under these conditions points toward a need for a better way to verify if we're seeing true impact, rather than just relying on the structure of the inferred kernel itself >

Warren: Even when they move to nonlinear or quasi-linear propagator estimation, it still reduces back to that same Wiener–Hopf deconvolution after some variable change. That means the non-identifiability isn't just a quirk of one specific model setup >

Iris: And they even use metaorder proxies reconstructed via daily segmentation in the SFK model to introduce an artificial long memory, which then triggers those universal decay theorems and confirms that agreement between an inferred propagator and a fine-tuned decay law isn't proof of transient impact >

Ruth: So, the authors conclude that the estimand TP identifies none of the underlying mechanisms, nor does it identify impact decay itself. The finding is that you can't use this specific estimation to tell if impact is decaying or not under these price conditions >

Warren: It sets up a diagnostic where you compare the inferred exponent against what's fixed by the input autocorrelation alone, and that comparison doesn't actually give you information about the underlying mechanism >

Iris: The inferred kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions to connect it to impact decay >

Ruth: It’s a lot of technical detail, but what it means for us in finance is that we have to be very careful about what we assume when we use these propagator models on data where prices are diffusive >

Warren: It suggests that relying solely on the inferred decay exponent as proof of impact decay is not reliable under long memory conditions, and you need more than just the exponent to make a causal claim >

Conclusion: Ruth: So we're wrapping up this look at the paper "Is price-impact decay statistically identifiable by propagator estimation? Diffusive prices force the inferred kernel to be transient and fine-tuned". Basically, they’re showing that for diffusive prices, you can’t tell if there’s real impact decay just by looking at how these propagator models estimate things.

Warren: That means the specific way we use this estimation tool doesn't actually reveal whether market impact is decaying or not. It just spits out some kind of artificial decay pattern because of how the math works with diffusive price dynamics.

Iris: The authors use universal theorems from Wiener–Hopf deconvolution to prove that for these price types, the inferred kernel has to be transient anyway, and its decay rate is set by the order flow autocorrelation alone.

Ruth: So what they’re saying is that if you see a certain fine-tuned decay in your estimates, it isn't proof of true impact decay because that pattern is baked into the deconvolution process itself.

Warren: That suggests if we look at these models, we have to be careful not to mistake an artifact of the estimation method for a real market mechanism. It’s about what the tool is actually measuring versus what it’s creating.

Iris: The paper shows that even in a model where impact is strictly permanent and doesn't decay, like their SFK model, the inferred propagator still shows that same artificial transient behavior. That confirms this issue holds up under different assumptions about impact.

Ruth: The implication here is pretty big for how we try to study market structure using these models. It suggests that just matching an estimated decay curve to a theoretical one isn't enough; you need a way to check if the estimation method itself is robust against things like model misspecification.

Warren: So the takeaway is that this specific estimation technique doesn’t give us direct evidence about impact decay under these conditions. It points us toward needing different ways to identify those underlying mechanisms if we want causal conclusions.

Shunta Fujiwara, Yuki Sato, Kiyoshi Kanazawa

Kyoto University · University of Tsukuba

q-fin.TR, cond-mat.stat-mech, physics.soc-ph, q-fin.MF, q-fin.ST

Submitted: 2026-11-07

Updated: 2026-11-07

Comments: Main text: 5 pages, 3 figures; End Matter and Supplementary Material: 49 pages

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 77/100

The gist: Is price-impact decay statistically identifiable by propagator estimation? This paper proves that asymptotic impact decay is not statistically identifiable by propagator estimation because for…

Key concepts

Propagator Estimation
This involves estimating a kernel (a function describing how past order signs affect future prices) using mathematical tools. The paper shows that when prices are diffusive, this estimation method yields results that are not robust enough to distinguish between different types of price dynamics, such as true impact decay versus artificial decay.
Wiener–Hopf Deconvolution
This is a universal mathematical framework used in the estimation process. The paper relies on theorems from this field, which state that the resulting estimated kernel (propagator) must be transient. Its decay rate is fixed solely by the order-flow autocorrelation, regardless of whether true impact decay exists.
Diffusive Prices
This refers to a specific class of price dynamics where price movements are characterized by randomness or diffusion. The paper demonstrates that within this class, the estimation technique cannot distinguish between models with true asymptotic impact decay and those that do not, because the estimation process itself introduces an artificial transient behavior.
Fine-tuned Decay
This refers to a specific type of decay pattern observed in the estimated kernel. The paper argues that this fine-tuned decay is not a genuine feature of the underlying price impact mechanism; instead, it is an artifact introduced by the deconvolution estimation process when applied to diffusive price dynamics.

Terminology

Summary

Is price-impact decay statistically identifiable by propagator estimation? p. 0 This paper proves that asymptotic impact decay is not statistically identifiable by propagator estimation because for diffusive prices, the fine-tuned decay is returned whether or not the true impact decays

The Core Argument

The central finding is a no-go theorem demonstrating that asymptotic impact decay cannot be identified from propagator estimation when prices are diffusive because for such processes, the fine-tuned decay is an artifact of the deconvolution rather than evidence of true impact decay This result rests on universal theorems for Wiener–Hopf deconvolution which state that the inferred propagator is necessarily transient, and its decay exponent is fixed by the order-flow autocorrelation alone, irrespective of the true impact The authors demonstrate this using an exactly solvable null model where impact is strictly permanent, yet the inferred propagator shows an artificial fine-tuned decay

Universal Decay Theorems

The paper establishes universal decay theorems for the deconvolution estimation under diffusive price dynamics and long-memory order flow, stating that the inferred kernel is necessarily transient with an asymptotic form of either G(τ) = AGτ−βc + o(τ−βc) or G(τ) = AGτ−βc ln τ + o(τ−βc ln τ), where βc is fixed by the order-flow autocorrelation alone These theorems apply even to a null model with strictly permanent impact and no decay, such as the Sato–Fujiwara–Kanazawa (SFK) model The theorems show that the property of fine-tuned decay is a property of the deconvolution, not evidence of impact decay

Model Class Equivalence

The non-identifiability arises because the estimand TP = (G∞, β) depends only on γ within the diffusive class MD, meaning it carries no information beyond γ for any prior on MD The study shows that the linear propagator class and models outside this class, such as the SFK model with strictly permanent impact, are observationally equivalent with respect to TP because both map to inferred kernels with G∞ = 0 and the same decay exponent βc This fragility under model misspecification suggests that the inferred decay has not been subjected to a robustness check against generative models outside the propagator class

Nonlinear Estimation and Proxies

The non-identifiability persists even when moving to nonlinear (quasi-linear) propagator estimation because it reduces to the same Wiener–Hopf deconvolution after a change of variables Furthermore, using metaorder proxies reconstructed via daily segmentation in the SFK model introduces an artificial long memory, which causes the universal decay theorems to apply and confirm that agreement between an inferred propagator and fine-tuned decay law cannot be taken as evidence for transient impact

Conclusion on Identification

The analysis confirms that the estimand TP = (G∞, β) identifies none of the underlying mechanisms, nor impact decay itself The conclusion is that agreement between an inferred propagator G(τ) and a fine-tuned decay law cannot be taken as evidence for transient impact This suggests a diagnostic where the inferred exponent is compared with the value fixed by the input autocorrelation alone, and this agreement carries no information about the mechanism The inferred kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions

SFK Model Validation

The numerical validation using the SFK model confirms that even though it has no true decay, its inferred kernel still shows an artificial fine-tuned decay G(τ) ∝ τ−βc This confirms the prediction of universal decay theorems for the SFK model The analysis of the inferred propagator Gn shows that G∞ = 0, implying that the inferred propagator is transient The coefficient AG is shown to be non-zero, which implies AG = B(βc, γ) sin(πβc)/ (2 AS AC) > 0

Asymptotic Behavior of Gn

The asymptotic behavior of the inferred propagator Gn is characterized by limn→∞ n βc ln n Gn = AG, where AG is given by B(βc, γ) sin(πβc)/ (2 AS AC) The analysis splits Gn into three parts: G1(n), G2(n), and G3(n), showing that only the leading term G1(n) converges to AG, while the lower-order terms G2(n) and G3(n) vanish as n → ∞ This confirms that the inferred propagator Gn is transient

Final Identification Diagnostic

The paper suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone This comparison carries no information about the mechanism, serving as a prompt to test the estimation outside the fitted class The inferred fine-tuned decay is a property of the estimator before it is a finding about the dynamics The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimator TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself The universality theorem is the mirror image of a no-go theorem for identifying microscopic dynamics from the universal observables alone In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel remains the best linear predictor of price changes from past order signs, but causal reading requires additional identification assumptions In summary, under diffusivity and long memory, the fine-tuned decay is a property of the estimator before it is a finding about the dynamics The estimand TP = (G∞, β) identifies none of these mechanisms, nor impact decay itself This suggests a diagnostic where agreement between an inferred propagator and a fine-tuned decay law is tested by comparing the inferred exponent with the value fixed by the input autocorrelation alone The kernel …

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved identification of market mechanisms by using propagator estimation as a diagnostic tool to distinguish between different price formation models, such as those with transient impact versus those with permanent impact, by comparing inferred decay exponents to values fixed solely by order-flow autocorrelation.

  2. Enhanced model selection capabilities where an AI system can assess the observational equivalence of different classes of processes, like the transient linear propagator class and the Sato–Fujiwara–Kanazawa model, based on whether their inferred kernels share the same asymptotic behavior for a given estimand.

  3. Robustness checking for estimation artifacts by implementing null models, specifically using an exactly solvable null model with strictly permanent impact to test whether an observed fine-tuned decay in propagator estimation is a genuine property of the data-generating process or merely an artifact of the Wiener–Hopf deconvolution itself.

  4. Advanced analysis of nonlinear propagator estimation by recognizing that quasi-linear estimators reduce to the same deconvolution, allowing the AI system to predict identification failure when true metaorders have long memory, even if they are modeled using daily segmentation proxies that introduce artificial correlation.

  5. Accurate asymptotic prediction of inferred kernel behavior by utilizing the universal decay theorems to determine whether price dynamics fall into three distinct regimes: transient decay with a fine-tuned law, logarithmic correction at the margin, or a strictly permanent impact scenario.

Sources

Related papers