关于能在实盘市场中经受检验的交易系统的研究。https://marketmaker.cc/zhhttps://marketmaker.cc/zh/blog/arbitrage-prereq-6-rust-basics/https://marketmaker.cc/zh/blog/arbitrage-prereq-6-rust-basics/套利课程的从零开始短课。通过一个读取四个价格的小程序，讲解 Rust 的变量、循环、向量和函数，以及为什么发布版本更快。代码可运行，附有输出。Thu, 12 Nov 2026 00:00:00 GMT套利课程预备课rust性能https://marketmaker.cc/zh/blog/arbitrage-prereq-5-probability/https://marketmaker.cc/zh/blog/arbitrage-prereq-5-probability/套利课程的从零开始短课。什么是概率，怎样通过计数来测量它，什么是钟形曲线，以及为什么 z 分数大于 2 很少见。代码可运行，附有输出。Thu, 12 Nov 2026 00:00:00 GMT套利课程预备课概率正态分布Pythonhttps://marketmaker.cc/zh/blog/arbitrage-prereq-4-vectors-matrices/https://marketmaker.cc/zh/blog/arbitrage-prereq-4-vectors-matrices/套利课程的从零开始短课。把一列数看作向量，把表格看作矩阵，用 numpy 对所有数一次完成同一个运算，以及什么是广播。代码可运行，附有输出。Thu, 12 Nov 2026 00:00:00 GMT套利课程预备课numpy向量矩阵Pythonhttps://marketmaker.cc/zh/blog/arbitrage-prereq-3-mean-stdev/https://marketmaker.cc/zh/blog/arbitrage-prereq-3-mean-stdev/套利课程的短篇零基础课。在十个数字上讲解均值、标准差和 z-score，说明为什么一个异常值会改变一切，以及 z-score 如何告诉你何时出现异常。Mon, 09 Nov 2026 00:00:00 GMT套利课程先修知识统计z-scorePythonhttps://marketmaker.cc/zh/blog/arbitrage-prereq-2-graphs/https://marketmaker.cc/zh/blog/arbitrage-prereq-2-graphs/套利课程的短篇零基础课。什么是图，如何用 Python 存储图，如何列出两点之间的每条路线，以及为什么地图变大后，逐条尝试路线会失效。Mon, 09 Nov 2026 00:00:00 GMT套利课程先修知识图最短路径Pythonhttps://marketmaker.cc/zh/blog/arbitrage-prereq-1-logarithms/https://marketmaker.cc/zh/blog/arbitrage-prereq-1-logarithms/套利课程的短篇零基础课。对数如何处理一串汇率，为什么对数之和取代乘积，以及如何理解负号。附带可手动核对输出的简短 Python 代码。Mon, 09 Nov 2026 00:00:00 GMT套利课程先修知识对数数学Pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-16-release/https://marketmaker.cc/zh/blog/arbitrage-course-16-release/套利课程第 16 课，也是最后一课。将纸面交易机器人变成可以展示的成果：交易日志、报告页面、README、没有秘密信息的干净公开仓库，以及诚实介绍结果的方法。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者发布READMEgithub纸面交易pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-15-pairs-trading/https://marketmaker.cc/zh/blog/arbitrage-course-15-pairs-trading/套利课程第 15 课。为什么比较两个原始价格没有意义，价差及其 z-score 如何显示两枚关联币何时偏离，以及一个展示手续费如何吃掉优势的小型回测。附有可复现输出的 Python 代码。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者配对交易统计套利z-score均值回归pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-14-futures-spot-funding/https://marketmaker.cc/zh/blog/arbitrage-course-14-futures-spot-funding/套利课程第 14 课。什么是永续期货及其资金费率，如何在现货市场持有一枚币、在期货上持有相反仓位，以及手续费、负费率和强平如何影响结果。附有可手动核对数字的简短 Python 代码。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者期货资金费率永续合约现货持有套利pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-13-speed/https://marketmaker.cc/zh/blog/arbitrage-course-13-speed/套利课程第 13 课。用循环检查每个三币种环路，再用 numpy 矩阵技巧和一段简短的 Rust 程序检查，并测量三种方法。附有可运行的代码和实际计时结果。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者numpy向量化rust性能pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-12-paper-trading-risk/https://marketmaker.cc/zh/blog/arbitrage-course-12-paper-trading-risk/套利课程第 12 课。让机器人使用实时价格和虚拟资金运行，将每个数字移入配置文件，并加入风险限额，在糟糕的一小时造成损失前停止机器人。附带可核对输出的简短 Python 代码。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者模拟交易风险管理配置Pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-11-reality-check/https://marketmaker.cc/zh/blog/arbitrage-course-11-reality-check/套利课程第 11 课。了解延迟、部分成交、第二条交易腿失败和过时价格，并运行一个小型模拟，看看盈利计划如何变成平均亏损。附有可运行和修改的简短 Python 代码。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者延迟执行风险模拟pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-10-size-fees-slippage/https://marketmaker.cc/zh/blog/arbitrage-course-10-size-fees-slippage/套利课程第 10 课。结合手续费和订单簿遍历，计算不同交易数量下的利润，并找到利润最高的数量。附带可手工核对数字的简短 Python 代码。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者手续费滑点仓位大小pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-09-order-book/https://marketmaker.cc/zh/blog/arbitrage-course-09-order-book/套利课程第 9 课。了解什么是订单簿，如何读取带数量的买单和卖单，如何逐档计算真实成交价格，并用一段可以手算验证的 Python 函数来练习。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者订单簿滑点pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-08-rich/https://marketmaker.cc/zh/blog/arbitrage-course-08-rich/套利课程第 8 课。了解机器人为什么需要限制步数的环、RICH 算法（VLDB 2025）背后的颜色编码技巧、一份可运行的小型 Python 实现，以及坦诚说明哪些内容做了简化。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者RICH颜色编码负环k 跳pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-07-spfa/https://marketmaker.cc/zh/blog/arbitrage-course-07-spfa/套利课程第 7 课。了解 SPFA 如何只重新检查可能发生变化的道路、一步步看队列示例、带负环警报的可运行 Python 代码，以及它与 Bellman-Ford 的速度对比。Mon, 09 Nov 2026 00:00:00 GMT套利课程初学者SPFABellman-Ford队列负环pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-06-bellman-ford/https://marketmaker.cc/zh/blog/arbitrage-course-06-bellman-ford/套利课程第 6 课。通过一个小例子讲解 Bellman-Ford 算法，说明为什么多做一轮就能发现负环，并提供可返回盈利兑换环的 Python 代码，以及手动检查的方法。Mon, 09 Nov 2026 00:00:00 GMT套利课程入门Bellman-Ford负环图算法pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-05-rates-as-a-graph/https://marketmaker.cc/zh/blog/arbitrage-course-05-rates-as-a-graph/套利课程第 5 课。把货币看作点、把汇率看作箭头，了解为什么取对数后乘法会变成加法，以及为什么有利可图的兑换环会变成总和为负的环。Mon, 09 Nov 2026 00:00:00 GMT套利课程入门图对数三角套利https://marketmaker.cc/zh/blog/arbitrage-course-04-compare-best-prices/https://marketmaker.cc/zh/blog/arbitrage-course-04-compare-best-prices/套利课程第 4 课。价差、手续费、报价货币和单位，以及一个简短的 Python 程序，用于扫描五家交易所并报告扣除手续费后仍然存在的价差。Mon, 09 Nov 2026 00:00:00 GMT套利课程入门价差手续费pythonhttps://marketmaker.cc/zh/blog/arbitrage-course-03-stages-and-design/https://marketmaker.cc/zh/blog/arbitrage-course-03-stages-and-design/套利课程第3课。了解机器人为何分阶段工作、每个阶段负责什么，以及如何让 AI 智能体提供一份你能读懂和检查的设计方案。Mon, 09 Nov 2026 00:00:00 GMT套利课程入门架构订单簿模拟交易https://marketmaker.cc/zh/blog/arbitrage-course-02-kinds-of-arbitrage/https://marketmaker.cc/zh/blog/arbitrage-course-02-kinds-of-arbitrage/套利课程第2课。了解四种主要套利类型（跨交易所、循环、统计套利、期现套利），每种类型的一个小例子，以及如何选择机器人要寻找的机会。Mon, 09 Nov 2026 00:00:00 GMT套利课程入门三角套利统计套利资金费率https://marketmaker.cc/zh/blog/arbitrage-course-01-what-is-arbitrage/https://marketmaker.cc/zh/blog/arbitrage-course-01-what-is-arbitrage/套利课程第1课。了解套利的含义、价格差为何存在，并用十行 Python 代码编写一个扫描器，比较两家交易所同一种币的最佳买价和最佳卖价。Mon, 09 Nov 2026 00:00:00 GMT套利课程入门买价卖价pythonhttps://marketmaker.cc/zh/blog/xgboost-gradient-boosting-trading/https://marketmaker.cc/zh/blog/xgboost-gradient-boosting-trading/收益方向分类器属于不平衡问题，0.5 不是正确的决策阈值。本文在加密货币数据上比较 scale_pos_weight、焦点损失和精度约束阈值优化，并介绍 XGBoost、LightGBM 与 CatBoost 之间的工程差异。Tue, 25 Aug 2026 00:00:00 GMTmachine-learningxgboostgradient-boostingquantpredictionhttps://marketmaker.cc/zh/blog/transfer-entropy-crypto-information-flow/https://marketmaker.cc/zh/blog/transfer-entropy-crypto-information-flow/DCC-GARCH tells you when crypto dependence tightens. Transfer entropy tells you which way it points. A directed information-flow measure, its null calibration, and an honest account of what it does and does not add over average pairwise correlation.Mon, 24 Aug 2026 00:00:00 GMTcausal-inferencetransfer-entropyinformation-theorycryptonetworkhttps://marketmaker.cc/zh/blog/trade-classification-algorithms/https://marketmaker.cc/zh/blog/trade-classification-algorithms/Reconstructing trade direction from OHLCV, bar history, or venues without an aggressor flag - the classic rules, the stale-quote problem, bulk volume classification, and how to measure them against crypto's free ground truth.Sun, 23 Aug 2026 00:00:00 GMTmicrostructuretrade-classificationLee-Readyorder-flowdeep-learninghttps://marketmaker.cc/zh/blog/tick-data-forecasting-deep-learning/https://marketmaker.cc/zh/blog/tick-data-forecasting-deep-learning/Sequence models fed tick data still assume regular spacing. Three ways to tell a Transformer when a tick actually happened - learnable-timescale continuous encoding, ODE-RNN latent state, and delta_t as a plain feature - and the ablation that decides between them.Sat, 22 Aug 2026 00:00:00 GMTHFTtick-datadeep-learninghigh-frequencypredictionhttps://marketmaker.cc/zh/blog/synthetic-control-trading-evaluation/https://marketmaker.cc/zh/blog/synthetic-control-trading-evaluation/This series has priced the selection route to a false edge - DSR prices the winner, PBO prices the search. Neither touches confounding: the strategy that made money because volatility doubled the week you deployed it. The Synthetic Control Method builds a weighted counterfactual from a donor pool of untouched instruments and gives you a falsification criterion and a placebo p-value.Fri, 21 Aug 2026 00:00:00 GMTcausal-inferencesynthetic-controlcounterfactualevaluationeconometricshttps://marketmaker.cc/zh/blog/probabilistic-forecasting-trading/https://marketmaker.cc/zh/blog/probabilistic-forecasting-trading/How to evaluate a predictive distribution honestly - CRPS as a proper scoring rule, the PIT histogram as a calibration diagnostic, and DeepAR sampling in GluonTS.Thu, 20 Aug 2026 00:00:00 GMTforecastingprobabilisticquantile-regressionDeepARuncertaintyhttps://marketmaker.cc/zh/blog/pinn-options-pricing/https://marketmaker.cc/zh/blog/pinn-options-pricing/Putting the Heston, free-boundary, and jump-diffusion pricing PDEs into a neural network loss - the log-price residual, the mixed-partial autograd trick, and what still needs measuring.Wed, 19 Aug 2026 00:00:00 GMTdeep-learningPINNoptionsBlack-ScholesPDEhttps://marketmaker.cc/zh/blog/pcmci-causal-discovery-crypto/https://marketmaker.cc/zh/blog/pcmci-causal-discovery-crypto/How PCMCI's two-stage MCI test recovers directed causal links between crypto assets where correlation and bivariate Granger cannot - the construction, the tigramite pipeline, and the real-data study it still needs.Tue, 18 Aug 2026 00:00:00 GMTcausal-inferencePCMCIcausal-discoverytime-seriescryptohttps://marketmaker.cc/zh/blog/order-flow-imbalance-prediction/https://marketmaker.cc/zh/blog/order-flow-imbalance-prediction/Turning raw book updates into a signed flow quantity: the CKS event decomposition, multi-level OFI with PCA reduction, and Lee-Ready trade classification - plus an honest accounting of what the headline R-squared actually measures.Mon, 17 Aug 2026 00:00:00 GMTmicrostructureorder-flowimbalancepredictionHFThttps://marketmaker.cc/zh/blog/neural-ode-continuous-finance/https://marketmaker.cc/zh/blog/neural-ode-continuous-finance/Neural ODEs, Neural SDEs and continuous normalizing flows for irregularly-sampled market data - and the one ablation that decides whether continuous dynamics are worth their solver cost.Sun, 16 Aug 2026 00:00:00 GMTdeep-learningneural-ODEcontinuous-timeSDEdynamicshttps://marketmaker.cc/zh/blog/multi-task-learning-trading/https://marketmaker.cc/zh/blog/multi-task-learning-trading/Does jointly predicting return, volume, and volatility actually help? Measuring loss-balancing schemes and diagnosing negative transfer through gradient cosine similarity - with a classical baseline and purged walk-forward folds.Sat, 15 Aug 2026 00:00:00 GMTdeep-learningmulti-taskshared-representationpredictionauxiliaryhttps://marketmaker.cc/zh/blog/model-pruning-low-latency-trading/https://marketmaker.cc/zh/blog/model-pruning-low-latency-trading/Magnitude and structured pruning, the Lottery Ticket Hypothesis, movement pruning, distillation and 2:4 sparsity - the methods behind shrinking a trading model, and what still has to be measured before any of it ships.Fri, 14 Aug 2026 00:00:00 GMTmodel-compressionpruninglottery-ticketlatencydeploymenthttps://marketmaker.cc/zh/blog/liquidity-prediction-execution/https://marketmaker.cc/zh/blog/liquidity-prediction-execution/Our VWAP article shipped a static, weekly-refit volume curve and called the forecaster the weakest link. This is the follow-up: a Bayesian intraday updater run against the same 500-parent BTCUSDT harness, with the IS delta conditioned on realized curve error.Thu, 13 Aug 2026 00:00:00 GMTmicrostructureliquidityexecutionpredictionorder-bookhttps://marketmaker.cc/zh/blog/koopman-operator-market-dynamics/https://marketmaker.cc/zh/blog/koopman-operator-market-dynamics/Dynamic Mode Decomposition fits a linear operator to nonlinear market dynamics. The only question that matters: do the fitted modes persist from one window to the next, and does the rolling spectral radius lead realised volatility? Here is the measurement protocol and the code to run it.Wed, 12 Aug 2026 00:00:00 GMTmathematicsKoopmandynamical-systemsspectralpredictionhttps://marketmaker.cc/zh/blog/knowledge-distillation-trading-models/https://marketmaker.cc/zh/blog/knowledge-distillation-trading-models/The blog's standing answer to the accuracy-vs-latency tension is a two-stage fast/slow split. Distillation is a different answer: train one small model to mimic the ensemble. The KD loss, temperature, born-again nets, early exits for a variable latency budget, and the distill-to-FPGA pipeline - plus the measurements that would decide whether it beats the two-stage split.Tue, 11 Aug 2026 00:00:00 GMTmodel-compressiondistillationlatencyHFTdeploymenthttps://marketmaker.cc/zh/blog/instrumental-variables-quantitative-finance/https://marketmaker.cc/zh/blog/instrumental-variables-quantitative-finance/Selection bias in the search is not the only way a regression fools you. When the regressor is correlated with the error, more data makes the estimate more confidently wrong. Instrumental variables applied to the one endogeneity problem this blog has already left open: permanent impact from net taker flow.Mon, 10 Aug 2026 00:00:00 GMTcausal-inferenceinstrumental-variables2SLSeconometricsendogeneityhttps://marketmaker.cc/zh/blog/hawkes-process-order-arrival/https://marketmaker.cc/zh/blog/hawkes-process-order-arrival/Fitting a self-exciting point process to real crypto trade tape: where the three numbers (mu, alpha, beta) come from, how to estimate the branching ratio n, whether the exponential kernel survives a goodness-of-fit test, and how much n moves when you change the estimation window.Sun, 09 Aug 2026 00:00:00 GMTmicrostructureHawkes-processpoint-processorder-arrivalcryptoHFThttps://marketmaker.cc/zh/blog/hamiltonian-neural-networks-finance/https://marketmaker.cc/zh/blog/hamiltonian-neural-networks-finance/Hamiltonian Neural Networks are provably stable - but stability is worthless if the conserved quantity does not exist. Learning a scalar H via autograd, symplectic integration, and the falsification test that decides whether a financial (q, p) pair is canonical at all.Sat, 08 Aug 2026 00:00:00 GMTdeep-learningHamiltonianphysics-informeddynamicsconservationhttps://marketmaker.cc/zh/blog/granger-causality-lead-lag-crypto/https://marketmaker.cc/zh/blog/granger-causality-lead-lag-crypto/Granger causality on crypto prices done two ways - differenced returns and Toda-Yamamoto on levels - with a correct Wald implementation, an effective-N corrected causality matrix, and a rolling-stability test of whether the lag is tradeable at all.Fri, 07 Aug 2026 00:00:00 GMTcausal-inferenceGranger-causalitylead-lagcryptoVARhttps://marketmaker.cc/zh/blog/gaussian-processes-trading/https://marketmaker.cc/zh/blog/gaussian-processes-trading/金融时间序列的核设计 - - 母体粗糙度、局部周期成分、光谱混合 - - 以及作为不需要验证集的正则化器的边际似然。加上在任何东西可以交易之前仍然需要衡量的诚实清单。Thu, 06 Aug 2026 00:00:00 GMTbayesiangaussian-processkerneluncertaintynon-parametrichttps://marketmaker.cc/zh/blog/fourier-neural-operator-finance/https://marketmaker.cc/zh/blog/fourier-neural-operator-finance/算子学习映射整个函数空间，而不是点 - - 傅里叶神经算子如何在频率空间中参数化 PDE 解算子，为期权定价购买什么，以及它的哪些主张仍然需要在真实硬件上测量。Wed, 05 Aug 2026 00:00:00 GMTdeep-learningFNOPDEoperator-learningoptionshttps://marketmaker.cc/zh/blog/flash-attention-trading-models/https://marketmaker.cc/zh/blog/flash-attention-trading-models/Flash Attention 使 23,400 步的交易上下文无需计算 - - 平铺、在线 softmax 和 N^2 d^2 / M 的 IO 界限。更长的上下文是否使模型更好是一个单独的问题，它必须被测量，而不是断言。Tue, 04 Aug 2026 00:00:00 GMTdeep-learningattentionFlash-AttentionGPUoptimizationhttps://marketmaker.cc/zh/blog/ensemble-methods-trading/https://marketmaker.cc/zh/blog/ensemble-methods-trading/本博客未涵盖有关交易集成的三件事：为什么协方差项（而不是模型计数）在集成误差中占主导地位，堆栈如何作为具有折叠元特征的信号组合层工作，以及跨多个信号的周转净额是否真正降低了执行成本。Mon, 03 Aug 2026 00:00:00 GMTmachine-learningensemblebaggingstackingalphahttps://marketmaker.cc/zh/blog/double-ml-causal-trading/https://marketmaker.cc/zh/blog/double-ml-causal-trading/本博客至今所有模型回答'什么预测什么'。Double ML 回答'什么导致什么' - - 带有可辩护的标准误。部分线性模型、Neyman 正交性、order book 数据上的清理交叉拟合，以及为何有效 DML 置信区间仅对一个预先指定的问题成立的诚实说明。Sun, 02 Aug 2026 00:00:00 GMTcausal-inferencedouble-MLtreatment-effecteconometricsquanthttps://marketmaker.cc/zh/blog/causal-forests-treatment-effects-trading/https://marketmaker.cc/zh/blog/causal-forests-treatment-effects-trading/本博客每个回测估计条件平均值。因果森林估计条件处理效果 - tau(x) 而非 mu(x) - 具有诚实分裂、自适应核权重表示以及告诉您发现的异质性是否真实的校准测试。Sat, 01 Aug 2026 00:00:00 GMTcausal-inferencecausal-forestheterogeneous-effectseconometricstreatmenthttps://marketmaker.cc/zh/blog/bayesian-neural-networks-trading/https://marketmaker.cc/zh/blog/bayesian-neural-networks-trading/本博客的每个 position sizing 规则都把不确定性坍缩成单一纯量。MC Dropout 和深度集成将其拆分为模型无知与市场噪声 - - 这两者值得不同的 position size。Fri, 31 Jul 2026 00:00:00 GMTbayesianuncertaintyneural-networkriskposition-sizinghttps://marketmaker.cc/zh/blog/automl-systematic-trading/https://marketmaker.cc/zh/blog/automl-systematic-trading/自动特征生成（tsfresh、Featuretools）、预算感知的模型搜索（FLAML 的节约成本优化器）以及 WorldQuant 公式化 alpha 工厂 - - 这篇博客的搜索与过拟合系列尚未覆盖的研究流程部分，以及该系列自身结果对它们的启示。Thu, 30 Jul 2026 00:00:00 GMTAutoMLNASfeature-engineeringautomationquanthttps://marketmaker.cc/zh/blog/liquidation-cascades-trading-signal/https://marketmaker.cc/zh/blog/liquidation-cascades-trading-signal/Перевод статьи на zhWed, 29 Jul 2026 00:00:00 GMTliquidationscascadetrading signaldefiperpetualsorder bookmarket microstructurecrypto