Paper: arXiv 2305.13123
Authors: Matthieu Garcin
Abstract
We are interested in the nonparametric estimation of the probability density of price returns, using the kernel approach. The output of the method heavily relies on the selection of a bandwidth parameter. Many selection methods have been proposed in the statistical literature. We put forward an alternative selection method based on a criterion coming from information theory and from the physics of complex systems: the bandwidth to be selected maximizes a new measure of complexity, with the aim of avoiding both overfitting and underfitting. We review existing methods of bandwidth selection and show that they lead to contradictory conclusions regarding the complexity of the probability distribution of price returns. This has also some striking consequences in the evaluation of the relevance of the efficient market hypothesis. We apply these methods to real financial data, focusing on the Bitcoin.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced nonparametric statistics, information theory, and physics-inspired complexity measures, involving substantial mathematical derivations and kernel density estimation theory. It also applies these methods to real Bitcoin data, comparing bandwidth selection techniques and linking results to the efficient market hypothesis, demonstrating a strong empirical component.
Research Flowchart
flowchart TD
A["Research Goal: <br>Determine Optimal Bandwidth<br>for Kernel Density Estimation<br>of Bitcoin Returns"] --> B["Methodology: <br>Apply Multiple Bandwidth<br>Selection Methods"]
B --> C{"Key Methods Compared"}
C --> D["Proposed Method:<br>Complexity Maximization"]
C --> E["Existing Methods:<br>Plug-in, Cross-Validation"]
D & E --> F["Computation: <br>Estimate Probability Density<br>of Bitcoin Returns"]
F --> G["Key Outcomes"]
G --> H1["Complexity Method<br>Avoids Over/Underfitting"]
G --> H2["Existing Methods<br>Contradict Each Other"]
G --> H3["Implications for EMH:<br>Market Inefficiencies<br>are Detectable"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a new bandwidth selection method for kernel density estimation that maximizes a complexity measure tailored to financial data.
Source sentence
We put forward an alternative selection method based on a criterion coming from information theory and from the physics of complex systems: the bandwidth to be selected maximizes a new measure of complexity, with the aim of avoiding both overfitting and underfitting.
Assumes 6
- The observed price returns are independent and identically distributed (i.i.d.) samples. statistical
Source sentence
2.1 Kernel density estimation We assume we are given n independent and identically distributed (i.i.d.) observations X1, ..., Xn, of pdf f.
- A Gaussian kernel is used for the kernel density estimation. model
Source sentence
In Sections 3 and 4, we focus on the particular case of a Gaussian kernel.
- Bandwidths larger than a threshold hp lead to overly smoothed densities and are therefore excluded. model
Source sentence
(5) 2.4 A complexity-based method of bandwidth selection As exposed in Section 2.3, we focus on values of bandwidth in the interval (0, hp].
- The Gaussian distribution is taken as the low‑complexity reference parametric model for financial returns. model
Source sentence
Instead, the simple parametric distribution which is overused in finance is the Gaussian distribution.
Show 2 more
- The empirical distribution (the sample) represents the maximum‑disorder state (over‑fitting) as bandwidth h → 0. model
Source sentence
In this framework, we consider that the maximum disorder is the empirical distribution, towards which the kernel distribution tends when h →0.
- Bitcoin daily returns are treated as a stationary series for each one‑year window. data
Source sentence
In our analysis, we work with daily price returns and are interested in their kernel density estimation, limited to a one-year period, so that we get one estimated density for each of the 8 years in the time interval of the study.
Data
- Bitcoin daily price returns (BTC‑USD) crypto-onchain public
Source sentence
We have imported the data from Yahoo Finance. In our analysis, we work with daily price returns and are interested in their kernel density estimation, limited to a one-year period, so that we get one estimated density for each of the 8 years in the time interval of the study.
- Simulated datasets of size 1,000 synthetic public
Source sentence
The simulations are generated according either to a standard Gaussian distribution, or to a mixture of two Gaussian distributions 0.6g(x + 1.25) + 0.4g(x −1.25), where g is the standard Gaussian density, or to a fat-tailed Student distribution with 5 degrees of freedom.
Costs
Not applicable
Finds 3
- In simulations, the hPIT bandwidth is the smallest, hc (complexity‑based) and hAMISE are close, while hlik is the largest. statistical
Source sentence
For the Gaussian and the Student distributions, hPIT is the smallest bandwidth. Then, the values of hc and hAMISE are quite close to each other, whereas hlik is the largest bandwidth.
- The complexity measure rises from zero as h → 0, reaches a clear local maximum at the complexity‑based bandwidth hc, then declines toward hp. theoretical
Source sentence
Starting to a value asymptotically equal to 0 when h →0, increasing h increases the complexity up to a very clear local maximum in hc, before a decrease until the bandwidth reaches the value of hp.
- Empirically, the choice of bandwidth (or other methodological choices) can lead the same Bitcoin dataset to be judged either consistent with the Efficient Market Hypothesis or not. statistical
Source sentence
The conclusion of this empirical study is that, depending on the bandwidth or on other methodological choices, a same dataset may either be considered as consistent with the EMH or not.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 8 related papers). Coverage is partial, so this is neither replication nor refutation.
2 papers in the archive
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.