微观运力驱动的集装箱运价预测：基于AIS数据与分解预测模型的实证研究

doi:10.16381/j.cnki.issn1003-207x.2024.1984

摘要/Abstract

摘要：

本文旨在研究集装箱航运市场微观信息对运价的影响机制并提升运价预测精度。首先，基于全球船舶自动识别系统轨迹数据构建营运运力指数，量化各航线实际投入运力，从微观层面映射供给端的运力动态调整与需求端的货流周期性波动。其次，采用季节性时序分解方法解析运力与运价的周期、趋势等特征，利用其经济学可解释性，从供需视角为运价的影响机制提供微观证据。最后，提出“分解-预测”框架并设计6种运力预测模型，同时，基于经验模态分解进行稳健性检验。实证结果表明：趋势项模型与全成分模型的预测效果最优，平均绝对百分比误差较自回归滑动平均模型基准降低约7个百分点；因季节性时序分解方法能有效提取季节性成分，其预测精度显著优于经验模态分解方法（差距达5个百分点）。本文从微观层面揭示了季节性需求波动与供给调整滞后共同驱动运价形成的内在机制，且考虑季节性的预测模型更适用于集装箱航运市场预测。

关键词: AIS数据, 营运运力, 运价, 集装箱航运市场, 分解-预测

Abstract:

Container shipping serves as a core pillar of global trade， bearing 52% of maritime trade value by worth. However， its freight rates exhibit extreme volatility - the China Containerized Freight Index （CCFI） surged by 215% year-on-year in August 2021， presenting severe challenges for shipping operations and derivative market risk management. Traditional research predominantly relies on macro-level indicators to analyze rate fluctuations but struggles to capture dynamic micro-level market heterogeneity. Although seasonal patterns have been identified by some scholars， their micro-level determinants lack empirical validation. Resolving how to analyze the dual supply-demand driving mechanisms through high-frequency microdata and enhance freight rate prediction accuracy has become an urgent scientific challenge. An operating capacity index is first constructed using massive AIS data， quantifying actual deployed capacity across shipping routes to precisely map supply-side capacity adjustments and demand-side cargo flow cyclicality at the micro level. Secondly， Seasonal-Trend decomposition using Loess （STL） is employed to decompose operating capacity and freight rates into seasonal， trend， and residual components. Leveraging the economic interpretability of these components， micro-level evidence of freight rate formation mechanisms from supply-demand perspectives is provided. Furthermore， a “decomposition-forecasting” approach is proposed. Through empirical comparisons of six capacity models （incorporating operating capacity and its decomposed components） against the ARMA（1，1） benchmark model， it is found that models considering only trend components or simultaneously incorporating all components achieve the highest prediction accuracy， with MAPE improving by 7 percentage points over the benchmark. Finally， robustness checks using Empirical Mode Decomposition （EMD） confirm the method's effectiveness， though its weaker seasonal component extraction capability results in 5 percentage point lower accuracy compared to STL. This micro-level evidence confirms that seasonal-incorporated prediction models are more suitable for container freight rate forecasting. The operating capacity index developed in this study addresses the limitations of traditional macro-indicators in capturing micro-market heterogeneity. By revealing dual supply-demand driving mechanisms through decomposition frameworks， it provides micro-empirical support for dynamic pricing models and offers decision-making foundations for shipping companies' capacity allocation and derivative pricing （e.g.， freight futures）.

Key words: AIS data, operating capacity, freight rates, container shipping market, decomposition-prediction

中图分类号:

隋聪,王尚,梁靖敏, 等. 微观运力驱动的集装箱运价预测：基于AIS数据与分解预测模型的实证研究[J]. 中国管理科学, 2026, 34(8): 38-48.

Cong Sui,Shang Wang,Jingmin Liang, et al. Micro-Level Capacity-Driven Container Freight Rate Forecasting: An Empirical Study Using AIS Data and Decomposition-Based Prediction Models[J]. Chinese Journal of Management Science, 2026, 34(8): 38-48.

图/表 11

表1

营运运力指数样本航线"

航线	营运运力	计量单位	基本港/基本国
日本航线	$C a p 1$	TEUs	日本
欧洲航线	$C a p 2$	TEUs	汉堡、鹿特丹、安特卫普、弗利克斯托、勒阿弗尔等
美西航线	$C a p 3$	TEUs	洛杉矶、长滩、西雅图、奥克兰、塔科马等
美东航线	$C a p 4$	TEUs	纽约-新泽西、萨凡纳、休斯顿、查尔斯顿、莫比尔等
韩国航线	$C a p 5$	TEUs	韩国
东南亚航线	$C a p 6$	TEUs	新加坡、越南、马来西亚、泰国、菲律宾、印度尼西亚
澳新航线	$C a p 7$	TEUs	澳大利亚、新西兰
南非航线	$C a p 8$	TEUs	南非
南美航线	$C a p 9$	TEUs	巴西、哥伦比亚、秘鲁、智利、厄瓜多尔等

表1

图1

图2

表2

描述性统计"

变量	数量	均值	标准差	偏度	峰度	ADF检验
$F r e$	575	0.208	0.252	1.934	2.416	0.191
$F r e S$	575	0.461	0.123	0.388	4.333	0.000
$F r e T$	575	0.219	0.275	1.733	1.756	0.022
$F r e R$	575	0.435	0.156	1.063	3.249	0.002
$C a p$	575	0.612	0.138	-0.948	1.997	0.000
$C a p S$	575	0.699	0.124	-1.057	2.888	0.000
$C a p T$	575	0.377	0.182	0.442	0.631	0.960
$C a p R$	575	0.591	0.125	-0.689	3.093	0.000
$Δ F r e$	575	0.516	0.188	0.655	0.303	0.013
$Δ C a p T$	575	0.503	0.279	-0.063	-1.123	0.046
$Δ C a p$	575	0.532	0.105	-0.406	3.640	0.000

表2

表3

参考文献 41

[1]	Chen R， Dong J X， Lee C Y. Pricing and competition in a shipping market with waste shipments and empty container repositioning［J］. Transportation Research Part B： Methodological， 2016， 85： 32-55.
[2]	Kilian L， Nomikos N K， Zhou X. Container Trade and the US Recovery［J］. International Journal of Central Banking， 2023， 19（1）： 417-450.
[3]	Choi T M， Chung S H， Zhuo X. Pricing with risk sensitive competing container shipping lines： Will risk seeking do more good than harm？［J］. Transportation Research Part B： Methodological， 2020， 133： 210-229.
[4]	Shi J， Jiao Y， Chen J， et al. Construction of resilience mechanisms in response to container shipping market volatility during the pandemic period： From the perspective of market supervision［J］. Ocean Coastal Management， 2023， 240： 106642.
[5]	Yu F， Xiang Z， Wang X， et al. An innovative tool for cost control under fragmented scenarios： The container freight index microinsurance［J］. Transportation Research Part E： Logistics and Transportation Review， 2023， 169： 102975.
[6]	Tvedt J， Hovi I B. Container shipping： A market equilibrium perspective on freight rates formation post-Covid-19［J］. Transportation Research Part A： Policy and Practice， 2024， 179： 103917.
[7]	隋聪，赵越，孙晓倩，等. 航运与大宗商品的跨市场影响——来自铁矿石及其航线的证据［J］. 系统工程理论与实践， 2022， 42（3）： 713-723.
	Sui C， Zhao Y， Sun X Q， et al. Cross-market impacts of shipping and bulk commodities： The evidence from iron ore and its routes［J］. Systems Engineering-Theory Practice， 2022， 42（3）： 713-723.
[8]	Luo M， Fan L， Liu L. An econometric analysis for container shipping market［J］. Maritime Policy Management， 2009， 36（6）： 507-523.
[9]	Michail N A， Melas K D. Shipping markets in turmoil： An analysis of the Covid-19 outbreak and its implications［J］. Transportation Research Interdisciplinary Perspectives， 2020， 7： 100178.
[10]	Sui C， Wang S， Zheng W. Sentiment as a shipping market predictor： Testing market-specific language models［J］. Transportation Research Part E： Logistics and Transportation Review， 2024， 189： 103651.
[11]	汤霞，匡海波，孟斌，等. 基于EMD的中国出口集装箱运价指数波动特性［J］. 科研管理， 2017， 38（12）： 144-154.
	Tang X， Kuang H B， Meng B， et al. Dynamic volatility of China’s containerised freight index based on EMD［J］. Science Research Management， 2017， 38（12）： 144-154.
[12]	Yin J， Shi J. Seasonality patterns in the container shipping freight rate market［J］. Maritime Policy Management， 2018， 45（2）： 159-173.
[13]	Regli F， Nomikos N K. The eye in the sky–Freight rate effects of tanker supply［J］. Transportation Research Part E： Logistics and Transportation Review， 2019， 125： 402-424.
[14]	Yu H， Fang Z， Lu F， et al. Impact of oil price fluctuations on tanker maritime network structure and traffic flow changes［J］.Applied Energy，2019，237： 390-403.
[15]	Zhang L， Meng Q， Fang Fwa T. Big AIS data based spatial-temporal analyses of ship traffic in Singapore port waters［J］. Transportation Research Part E： Logistics and Transportation Review， 2019， 129： 287-304.
[16]	Jin L， Chen J， Chen Z， et al. Impact of COVID-19 on China’s international liner shipping network based on AIS data［J］. Transport Policy， 2022， 121： 90-99.
[17]	Bai X， Li Y. The Congestion effect of oil transportation and its trade implications［R］. SSRN Working Paper，Social Science Research Network， 2021.
[18]	Elmaghraby W， Keskinocak P. Dynamic pricing in the presence of inventory considerations： Research overview， current practices， and future directions［J］. Management Science， 2003， 49（10）： 1287-1309.
[19]	Barberis N， Greenwood R， Jin L， et al. Extrapolation and bubbles［J］. Journal of Financial Economics， 2018， 129（2）： 203-227.
[20]	Moutzouris I C， Nomikos N K. The formation of forward freight agreement rates in dry bulk shipping： Spot rates， risk premia， and heterogeneous expectations［J］. Journal of Futures Markets， 2019， 39（8）： 1008-1031.
[21]	Vemuri S， Munim Z H. Seasonality and forecasting analysis of the South-East Asian container freight market［J］. Maritime Business Review， 2023， 8（2）： 121-138.
[22]	Munim Z H. State-space TBATS model for container freight rate forecasting with improved accuracy［J］. Maritime Transport Research， 2022， 3： 100057.
[23]	Munim Z H， Schramm H J. Forecasting container freight rates for major trade routes： A comparison of artificial neural networks and conventional models［J］. Maritime Economics Logistics， 2021， 23（2）： 310-327.
[24]	Cariou P， Guillotreau P. Capacity management by global shipping alliances： Findings from a game experiment［J］. Maritime Economics Logistics， 2022， 24（1）： 41-66.
[25]	Li L， Wan Y， Yang D. Do shipping alliances affect freight rates？ Evidence from global satellite ship data［J］. Transportation Research Part A： Policy and Practice， 2024， 181： 104010.
[26]	Zheng S， Wang K， Dong K， et al. Does the shipping alliance aggravate or alleviate container shipping market volatility［J］. Transportation Research Part A： Policy and Practice， 2024， 189： 104231.
[27]	Theodosiou M. Forecasting monthly and quarterly time series using STL decomposition［J］. International Journal of Forecasting， 2011， 27（4）： 1178-1195.
[28]	Ding L， Zhao H， Zhang R. Predicting multi-frequency crude oil price dynamics： Based on MIDAS and STL methods［J］. Energy， 2024， 313： 134003.
[29]	Yang T， Huang L， Fu P， et al. Combined multi-component composite time series power prediction model for distributed energy systems based on STL data decomposition［J］. Measurement， 2025， 253： 117299.
[30]	Trull O， García-Díaz J C， Peiró-Signes A. Multiple seasonal STL decomposition with discrete-interval moving seasonalities［J］. Applied Mathematics and Computation， 2022， 433： 127398.
[31]	Xu L， Ou Y， Cai J， et al. Offshore wind speed assessment with statistical and attention-based neural network methods based on STL decomposition［J］. Renewable Energy， 2023， 216： 119097.
[32]	Hyndman R J， Athanasopoulos G. Forecasting： principles and practice［M］. Melbourne： OTexts， 2018.
[33]	Bisoi R， Dash P K， Mishra S P. Modes decomposition method in fusion with robust random vector functional link network for crude oil price forecasting［J］. Applied Soft Computing， 2019， 80： 475-493.
[34]	Guo Y， Si J， Wang Y， et al. Ensemble-Empirical-Mode-Decomposition （EEMD） on SWH prediction： The effect of decomposed IMFs， continuous prediction duration， and data-driven models［J］. Ocean Engineering， 2025， 324： 120755.
[35]	Zhang X， Lai K K， Wang S Y. A new approach for crude oil price analysis based on Empirical Mode Decomposition［J］. Energy Economics， 2008， 30（3）： 905-918.
[36]	Yan Q， Wang S， Li B. Forecasting uranium resource price prediction by extreme learning machine with empirical mode decomposition and phase space reconstruction［J］. Discrete Dynamics in Nature and Society， 2014， 2014： 390579.
[37]	Yin K， Guo H， Yang W. A novel real-time multi-step forecasting system with a three-stage data preprocessing strategy for containerized freight market［J］. Expert Systems with Applications， 2024， 246： 123141.
[38]	Song W， Yang H， Li D， et al. Dynamic pricing and competition of container liner shipping services in a duopoly spot market［J］. Computers Industrial Engineering， 2023， 185： 109613.
[39]	Jiang H， Hu W， Xiao L， et al. A decomposition ensemble based deep learning approach for crude oil price forecasting［J］. Resources Policy， 2022， 78： 102855.
[40]	汤霞，匡海波，郭媛媛，等. 基于VMD的中国出口集装箱运价指数分析与组合预测［J］. 系统工程理论与实践， 2021， 41（1）： 176-187.
	Tang X， Kuang H B， Guo Y Y， et al. Analysis and combined forecasting of China containerized freight index based on VMD［J］. Systems Engineering-Theory Practice， 2021， 41（1）： 176-187.
[41]	Yu L， Wang Z， Tang L. A decomposition-ensemble model with data-characteristic-driven reconstruction for crude oil price forecasting［J］. Applied Energy， 2015， 156： 251-267.

变量	$Δ F r e$	$F r e T$	$F r e S$	$F r e R$	$Δ C a p$	$Δ C a p T$	$C a p S$	$C a p R$
$Δ F r e$	1.000
$F r e T$	0.006	1.000
$F r e S$	0.130^***	0.017	1.000
$F r e R$	-0.016	0.577^***	0.082^*	1.000
$Δ C a p$	-0.010	0.000	-0.016	-0.010	1.000
$Δ C a p T$	-0.366^***	0.077^*	-0.113^***	0.062	0.007	1.000
$C a p S$	0.124^***	0.001	0.198^***	-0.047	0.266^***	-0.015	1.000
$C a p R$	0.102^**	-0.059	-0.020	-0.128^***	0.567^***	-0.016	0.108^***	1.000

变量	(1)	(2)	(3)	(4)	(5)	(6)
$Δ C a p$		-0.007
$Δ C a p$		(-0.366)
$C a p S$			0.021			0.019
$C a p S$			(0.774)			(0.643)
$Δ C a p T$				-0.150^***		-0.149^***
$Δ C a p T$				(-4.213)		(-4.156)
$C a p R$					0.029	0.026
$C a p R$					(1.433)	(1.176)
$A R 1$	0.816^***	0.815^***	0.819^***	0.752^***	0.824^***	0.762^***
$A R 1$	(25.428)	(25.318)	(25.669)	(17.262)	(25.910)	(17.463)
$M A 1$	-0.318^***	-0.316^***	-0.328^***	-0.266^***	-0.337^***	-0.291^***
$M A 1$	(-5.972)	(-5.839)	(-6.068)	(-4.373)	(-6.288)	(-4.691)
$C$	0.584^***	0.588^***	0.569^***	0.651^***	0.567^***	0.623^***
$C$	(55.389)	(43.985)	(26.850)	(34.219)	(36.721)	(21.621)
$R 2$	0.426	0.426	0.426	0.437	0.428	0.439

变量	$F r e S$	$F r e T$	$F r e R$
$C a p S$	-0.004
$C a p S$	(-0.528)
$Δ C a p T$		-0.017^***
$Δ C a p T$		(-7.665)
$C a p R$			0.002
$C a p R$			(0.514)
$A R 1$	0.967^***	1.000^***	0.973^***
$A R 1$	(121.2)	(1817)	(165.4)
$M A 1$	0.306^***	0.996^***	0.356^***
$M A 1$	(11.986)	(59.246)	(13.484)
$C$	0.454^***	0.155	0.425^***
$C$	(11.694)	(0.268)	(9.335)
$R 2$	0.956	0.999	0.980

变量	模型1	模型2	模型3	模型4	模型5	模型6	模型7
Panel A 样本外时间区间为0.5年
$M A E$	0.089	0.089	0.087	0.064	0.089	0.063	0.492
$R M S E$	0.101	0.101	0.100	0.082	0.101	0.083	0.500
$M A P E$ (%)	20.012	20.012	19.560	13.458	19.938	13.237	98.573
Panel B 样本外时间区间为1.0年
$M A E$	0.124	0.124	0.123	0.091	0.124	0.091	0.486
$R M S E$	0.157	0.157	0.157	0.126	0.158	0.127	0.499
$M A P E$ (%)	26.669	26.670	26.599	18.876	26.811	19.031	100.218
Panel C 样本外时间区间为1.5年
$M A E$	0.101	0.101	0.101	0.078	0.101	0.078	0.517
$R M S E$	0.134	0.134	0.134	0.109	0.134	0.110	0.529
$M A P E$ (%)	21.104	21.102	21.098	15.657	21.129	15.779	100.201

变量	(1)	(2)	(3)	(4)	(5)	(6)
$Δ C a p$		-0.007
$Δ C a p$		(-0.366)
$C a p H$			0.034^*			0.034^*
$C a p H$			(1.737)			(1.666)
$C a p L$				-0.082^*		-0.085^*
$C a p L$				(-1.939)		(-1.959)
$C a p R$					-0.006	-0.014
$C a p R$					(-0.172)	(-0.406)
$A R 1$	0.816^***	0.815^***	0.827^***	0.797^***	0.816^***	0.807^***
$A R 1$	(25.428)	(25.318)	(26.353)	(22.868)	(25.329)	(22.956)
$M A 1$	-0.318^***	-0.316^***	-0.346^***	-0.303^***	-0.318^***	-0.331^***
$M A 1$	(-5.972)	(-5.839)	(-6.456)	(-5.439)	(-5.963)	(-5.765)
$C$	0.584^***	0.588^***	0.562^***	0.626^***	0.589^***	0.616
$C$	(55.389)	(43.985)	(34.943)	(26.118)	(21.590)	(14.845)
$R 2$	0.426	0.426	0.428	0.430	0.426	0.432