Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2004.01499

Deep Recurrent Modelling of Stationary Bitcoin Price Formation Using the Order Flow

Ye-Sheen Lim·Denise Gorse
View original source ↗Read source document ↗Export metadata ↓

Abstract

In this paper we propose a deep recurrent model based on the order flow for the stationary modelling of the high-frequency directional prices movements. The order flow is the microsecond stream of orders arriving at the exchange, driving the formation of prices seen on the price chart of a stock or currency. To test the stationarity of our proposed model we train our model on data before the 2017 Bitcoin bubble period and test our model during and after the bubble. We show that without any retraining, the proposed model is temporally stable even as Bitcoin trading shifts into an extremely volatile "bubble trouble" period. The significance of the result is shown by benchmarking against existing state-of-the-art models in the literature for modelling price formation using deep learning.

Explore related subjects

q-fin.STcs.LGq-fin.TRstat.ML

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Ye-Sheen Lim, Denise Gorse. 2020-03-31. Deep Recurrent Modelling of Stationary Bitcoin Price Formation Using the Order Flow. https://arxiv.org/abs/2004.01499

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Does Crypto Sentiment Extremity Widen Estimated Spreads? Evidence Depends on the Specification

We examine whether extreme values of the Crypto Fear & Greed Index are associated with a daily high-low spread estimate for Bitcoin. The sample contains 2,896 BTC/USDT observations from February 2018 to January 2026. We find an unconditional extreme-minus-neutral gap of 61.99 basis points. After close-to-close realised-volatility-quintile demeaning it is 24.79 basis points, although none of the five separate quintile contrasts survives Holm correction. With quadratic realised-volatility and strictly lagged momentum controls, the HAC estimate is 11.81 basis points (95% CI [-2.31,25.93], p=.101). A fixed non-parametric stratification gives 20.44 basis points (p=.0195 under circular shifts), while separate models for a zero-floored estimate's incidence and positive magnitude are imprecise. The results therefore show only a descriptive, specification-dependent association. We conclude that they do not establish a stable or causal liquidity premium.

q-fin.ST↗
arXiv2026

Do Cryptocurrency Markets Differentiate Infrastructure from Regulatory Shocks? A Multi-Moment Event Study with Dependence-Robust Inference

Do cryptocurrency markets respond differently to infrastructure and regulatory shocks? We study returns and conditional variance on a shared sample of 50 events and six assets (January 2019--August 2025), using GJR-GARCH-X models and dependence-aware inference. Treating event inclusion as a design parameter, we trace the variance differential across inclusion screens. Curated high-salience events yield a $3.49\times$ point-estimate multiplier, whereas a mechanical impact filter on a broad reconstructed candidate pool yields approximately $0.5$--$1.6\times$. This pattern is descriptive and selection-conditional, not an inferential comparison between screens. The curated variance differential is not significant against its fitted sharp per-asset-equality null under the conditional fixed-path Student-$t$-copula bootstrap ($p\approx0.39$). Floored recursive sensitivity gives one-sided $p=0.025$ at baseline and $0.041$ with asset-specific high-variance-regime controls, so the verdict depends on inference scheme and implementation. Across reported dependence inputs, the six-contrast effective sample size is $1.33$--$2.35$; one-sided effective-df sensitivity gives $p=0.044$--$0.116$. Earlier significance from treating correlated per-asset coefficients as independent samples is not robust to dependence and heavy-tail corrections. The cumulative-abnormal-return difference is $+8.69$ percentage points (event-level block-bootstrap $p=0.202$). The asymmetry remains directional, selection-conditional and unresolved. The contribution is an inference ladder and an internal Monte-Carlo calibration study, demonstrated through correction of the author's earlier significance claim.

q-fin.ST↗
arXiv2026

Modeling financial time series with $ϕ^{4}$ quantum field theory

We use a $ϕ^{4}$ quantum field theory with inhomogeneous couplings and explicit symmetry-breaking to model an ensemble of financial time series from the S$\&$P 500 index. The continuum nature of the $ϕ^4$ theory avoids the inaccuracies that occur in Ising-based models which require a discretization of the time series. We demonstrate this using the example of the 2008 global financial crisis. The $ϕ^{4}$ quantum field theory is expressive enough to reproduce the higher-order statistics such as the market kurtosis, which can serve as an indicator of possible market shocks. Accurate reproduction of high kurtosis is absent in binarized models. Therefore Ising models, despite being widely employed in econophysics, are incapable of fully representing empirical financial data, a limitation not present in the generalization of the $ϕ^{4}$ scalar field theory. We then investigate the scaling properties of the $ϕ^{4}$ machine learning algorithm and extract exponents which govern the behavior of the learned couplings (or weights and biases in ML language) in relation to the number of stocks in the model. Finally, we use our model to forecast the price changes of the AAPL, MSFT, and NVDA stocks. We conclude by discussing how the $ϕ^{4}$ scalar field theory could be used to build investment strategies and the possible intuitions that the QFT operations of dimensional compactification and renormalization can provide for financial modelling.

q-fin.ST↗