Paper: arXiv 2407.06745

Authors: Michael Sekatchev, Zhengxiang Zhou

Abstract

In this project, we propose to explore the Kalman filter’s performance for estimating asset prices. We begin by introducing a stochastic mean-reverting processes, the Ornstein-Uhlenbeck (OU) model. After this we discuss the Kalman filter in detail, and its application with this model. After a demonstration of the Kalman filter on a simulated OU process and a discussion of maximum likelihood estimation (MLE) for estimating model parameters, we apply the Kalman filter with the OU process and trailing parameter estimation to real stock market data. We finish by proposing a simple day-trading algorithm using the Kalman filter with the OU process and backtest its performance using Apple’s stock price. We then move to the Heston model, a combination of Geometric Brownian Motion and the OU process. Maximum likelihood estimation is commonly used for Heston model parameter estimation, which results in very complex forms. Here we propose an alternative but easier way of parameter estimation, called the method of moments (MOM). After the derivation of these estimators, we again apply this method to real stock data to assess its performance.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper involves advanced mathematics including stochastic differential equations, Ito calculus, and Kalman filter derivations, but also includes real-world data applications, backtesting, and algorithmic trading strategies.

Research Flowchart

  flowchart TD
  A["Research Goal: <br>Asset Price Analysis using Stochastic Models"] --> B["Data Inputs: <br>Simulated OU Process & Apple Stock Data"]
  B --> C["Model 1: Ornstein-Uhlenbeck (OU) <br>+ Kalman Filter"]
  C --> D{"Parameter Estimation <br>Method?"}
  D --> E["MLE (Simulation) <br>Trailing Est. (Real Data)"]
  E --> F["Outcome: <br>Day-Trading Algorithm & Backtest"]
  C --> F
  B --> G["Model 2: Heston Model <br>(GBM + OU Process)"]
  G --> H["Method of Moments (MOM) <br>for Parameter Estimation"]
  H --> I["Outcome: <br>Model Performance on Real Data"]

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 using a Kalman filter with an Ornstein‑Uhlenbeck process for asset price estimation, a recursive OU parameter estimation method, and a method‑of‑moments estimator for the Heston model.

Source sentence
In this project, we propose to explore the Kalman filter’s performance for estimating asset prices. ... we

Assumes 6

  • Asset returns follow an Ornstein‑Uhlenbeck process. model
    Source sentence
    filter will assume that the asset returns follow a Ornstein‑Uhlenbeck process, as is popular in the field.
  • Process and observation noises are independent and identically distributed (i.i.d.). statistical
    Source sentence
    In our applications, we will be assuming that the noise at each step is independent and identically distributed (i.i.d.), so we will be setting Qk = σ2 p1 for all k.
  • Observation noise is Gaussian. statistical
    Source sentence
    Suppose this true state Zk+1 is inaccessible, but it can be measured with some Gaussian noise, obtaining an observation ˜Zk+1...
  • All transaction costs are ignored in the trading simulation. execution
    Source sentence
    All transaction costs are neglected.
Show 2 more
  • Trading positions are opened at the market open price and closed at the market close price of the same day. execution
    Source sentence
    Based on the difference between the Kalman filter prediction and the true open price, a long or a short position is created with 100 shares of the stock
  • The optimal OU parameters evolve over time and can be estimated recursively using a fixed look‑back window tb. model
    Source sentence
    For this reason, we propose a recursive parameter estimation method.

Data

Backtest sample: Apple stock · equities · daily

  • Apple daily stock price (11‑month sample) market-prices public
    past eleven months · Apple Inc.
    Source sentence
    Figure 7: Apple’s daily stock price in the past eleven months with the Kalman filter overlaid.
  • Apple stock price (3‑year sample) market-prices public
    past 3 years · Apple Inc.
    Source sentence
    In this section, we aim to find the MOM estimators using (19) for Apple’s stock using a dataset that contains observations of price of the past 3 years.

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
All transaction costs are neglected.

Finds 4

Reported performance (Kalman filter day‑trading algorithm with OU model): annual return 12.2%

  • The optimal Kalman‑OU parameters for the three‑year backtest are σo = 3.36 and tb = 140 days. performance
    Source sentence
    The optimal pair of parameters from this plot is σo = 3.36, tb = 140 days.
  • Open‑price estimation using the recursive OU Kalman filter yields a very large sum of absolute errors (~38 453), about 50 per day. negative
    Source sentence
    the sum of absolute error is huge, which reaches around 38453. This means that the average daily absolute error is about 50.
  • Close‑price estimation yields an even larger sum of absolute errors (~90 959), about 120 per day. negative
    Source sentence
    the sum of absolute error is huge, which reaches around 90959. This means that the average daily absolute error is about 120.
  • The Kalman‑OU trading algorithm is compared against a buy‑and‑hold benchmark, with profits reported relative to the initial open price. performance
    Source sentence
    The performance of the algorithm is compared to a buy and hold strategy, which involves simply buying 100 shares of the stock at Zo0 and selling them at time T.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 16 related papers). Coverage is partial, so this is neither replication nor refutation.

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.

Cited References (11)

  1. Forecasting trends with asset prices (2017) — link
  2. Fast strong approximation Monte Carlo schemes for stochastic volatility models (2006) — link
  3. The Man Who Solved the Market: How Jim Simons Launched the Quant Revolution (2020)
  4. Theory of Speculation (1900)
  5. Estimation II (2001) — link
  6. Online Kalman Filter Tutorial — link
  7. A pairs trading strategy based on linear state space models and the Kalman filter (2016) — link
  8. Filtering Complex Turbulent Systems (2012) — link
and 3 more
  1. All of statistics: a concise course in statistical inference (2013)
  2. Maximum Likelihood Estimation of a Mean Reverting Process (2023) — link
  3. The Non-Misleading Value of Inferred Correlation: An Introduction to the Cointelation Model (2013) — link

0 of 11 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.