Method for constructing a steel cost forecast model in 2025: raw material price linkage and industry analysis technology

Method for constructing a steel cost forecast model in 2025: raw material price linkage and industry analysis technology

Against the backdrop of increasingly fierce competition in the steel industry, accurate prediction of steel costs has become a key means for companies to improve their profitability. This article will deeply analyze the core methods of building a steel cost model in 2025, combining the raw material price forecasting mechanism with industry linkage analysis technology to provide companies with a feasible cost control solution.

1. Steel cost composition and key driving factors
The cost of steel production is a complex dynamic system, which mainly includes three categories of expenses:

Raw material costs (accounting for 70%-75% of the total cost): Metal raw materials such as iron ore, coking coal, scrap steel and ferroalloys constitute the main cost of steel production. Data from January to May 2025 showed that the procurement cost of imported powder ore of the benchmarking company decreased by 15.66% year-on-year, while the cost of manganese ferroalloys rose by 0.69%, indicating that there are significant differences in price fluctuations of different raw material categories10.

Energy and manufacturing expenses (accounting for 15%-20%): covering electricity, gas consumption, equipment depreciation and environmental protection investment. As the "dual carbon" policy continues to advance, the environmental protection cost of steel companies has increased to 80-120 yuan per ton of steel in 2025, becoming an incremental factor that cannot be ignored.

Labor and operating costs (accounting for 8%-12%): including direct labor costs, logistics and management costs. It is worth noting that advanced companies have reduced labor costs per ton of steel by more than 30% through intelligent transformation.

*Table: Comparative analysis of the procurement costs of major raw materials from January to May 2025*

Raw material category Average cost (yuan/ton) Year-on-year change Price difference between leading and lagging enterprises
Imported powder ore 767.41 -15.66% 744~957 yuan (±14%)
Metallurgical coke 1546.15 -28.53% 1339~1798 yuan (±29%)
Scrap steel 2198.73 -14.46% 1924~2491 yuan (±26%)
Silicon manganese alloy 5532.21 -5.09% 5248~6174 yuan (±17%)
The core driving factors of cost differences are mainly reflected in three aspects: First, the procurement channel structure has become a key factor affecting costs. The cost of enterprises with self-owned coking plants is more than 200 yuan/ton lower than that of enterprises purchasing coke from outside, and the raw material stability of enterprises with long-term contract ore sources is significantly better than that of market procurement enterprises10. Secondly, the production process also directly affects the cost structure. Enterprises that use short-process steelmaking with electric furnaces have significant cost advantages when scrap steel prices are low, while the long process of blast furnace-converter is more competitive during the period of iron ore price decline. Finally, geographical location factors lead to obvious differences in logistics costs. The cost of imported ore to coastal steel mills is 120 yuan/ton lower than that of inland enterprises on average, while steel mills close to coal-producing areas can reduce fuel transportation costs by 15%.

2. Forecast of raw material price trends in 2025
①. Iron ore: Price decline under loose supply
In 2025, the global new iron ore production capacity is expected to reach 100 million tons, of which 30 million tons will be released in the second half of the year. Taking into account the natural attenuation of the four major mines, the net new production capacity is about 54 million tons. On the demand side, the newly started area of ​​domestic real estate is expected to decrease by 18% year-on-year, resulting in a 7%-9% decrease in total iron ore demand24. Under the change of supply and demand pattern, iron ore prices are expected to fluctuate and weaken, and the low point may reach 80 US dollars/dry ton (equivalent to about 580 yuan in RMB), and the average price for the whole year will remain in the range of 85-90 US dollars4.

②. Coking coal: The surplus pattern is difficult to reverse
In 2025, the coking coal market faces the pressure of oversupply: domestic production is expected to increase by 17 million tons, imports will decrease by 10 million tons, and the net increase in supply will be 7 million tons. On the demand side, the average daily molten iron output of 247 steel mills is 2.359 million tons, an increase of only 3.6% year-on-year, far lower than the supply growth rate2. In the first half of the year, the lowest coking coal futures price fell to 709 yuan/ton (equivalent to the level in 2016), and the price of Luliang low-sulfur main coking coal fell to 1,100 yuan/ton. In the second half of the year, the price is expected to fluctuate between 900-1,100 yuan/ton, and the low price zone will appear at the end of Q34.

③. Alloys and scrap steel: the trend of differentiation intensifies
Manganese alloys: Affected by the resumption of production in South African mines, the import volume of manganese ore has increased, but the operating rate of domestic silicon manganese plants has remained above 70%. The price range of silicon manganese is expected to be 5,000-5,500 yuan/ton, and ferromanganese is under pressure due to the reduction in stainless steel production
Scrap steel: The decline in the prosperity of the manufacturing industry has led to a decrease in processed scrap steel, but demolition scrap steel has increased with urban renewal. Changes in the supply structure have increased price volatility. The annual operating range is expected to be 1,900-2,300 yuan/ton

④. Policy and trade friction risks
The tariff exemption for Chinese steel products exported to the United States will expire in August 2025. If the negotiations fail, 2.5 million tons of steel will be transferred back to domestic sales, exacerbating the contradiction between market supply and demand4. At the same time, the EU Carbon Border Adjustment Mechanism (CBAM) is fully implemented, and the carbon cost of exported steel will increase by an additional 8-12 US dollars/ton8.

Table: Forecast range of main raw material prices in 2025

Raw material type Q1-Q2 actual average price Q3-Q4 forecast average price Full-year forecast center Key influencing factors
62% imported powder ore 767 yuan/ton 680-720 yuan/ton 710 yuan/ton Australian production capacity release progress
Metallurgical coke 1546 yuan/ton 1400-1500 yuan/ton 1450 yuan/ton Shanxi production reduction execution
Heavy scrap (≥6mm) 2198 yuan/ton 2050-2250 yuan/ton 2150 yuan/ton Demolition scrap steel increment ratio
Silicon manganese alloy 5532 yuan/ton 5300-5600 yuan/ton 5450 yuan/ton South African manganese ore arrival volume
3. Construction method of steel cost prediction model
①. Four-layer technical architecture design
The modern steel cost prediction model should adopt a layered architecture design to form a complete closed loop from data collection to decision support:

Data layer: Integrate multi-source heterogeneous data, including real-time raw material prices (iron ore, coking coal, alloys), energy costs (electricity prices, gas), logistics costs (sea transportation, land transportation) and policy texts (tariffs, environmental regulations). Data collection needs to cover 15 major global commodity exchanges, 8 port inventory databases and customs import and export data of various countries.

Algorithm layer: Adopt a multi-model fusion architecture, the core includes:

CNN convolutional neural network: Process unstructured data such as port inventory images and blast furnace operation heat maps to extract spatial features

Bi-LSTM bidirectional long short-term memory network: Analyze time series data and capture long-term price dependencies

Variational autoencoder (VAE): Generate synthetic samples in data missing scenarios to enhance model robustness3

Analysis layer: Introduce graph neural network (GNN) to build an industrial chain relationship map, quantify node impact (such as: the transmission intensity of coking plant shutdown to steel plant costs = 0.38), and identify key paths.

Application layer: output dynamic cost forecast, procurement strategy optimization, inventory warning threshold three types of decision support solutions, and display them in real time through visual dashboard.

②. Core algorithm combination application
In view of the particularity of steel cost forecast, a combined algorithm architecture should be adopted:

Cost-sensitive neural network: set higher weights for raw material costs, such as iron ore weight 0.35, coking coal 0.28, alloy 0.18, energy 0.12, and other 0.07, so that the model focuses more on key variables.

Multi-factor financial model: quantify interest rates, exchange rates, and tariff policies as follows:

text
Cost adjustment coefficient = 0.12×exchange rate fluctuation + 0.08×interest rate change + 0.25×tariff rate
Achieve macro factor embedding.

System dynamics simulation: Build a cost feedback system containing 200+ variables to simulate chain reactions such as "iron ore price increase → electric furnace steel ratio increase → scrap steel demand increase → arc furnace steel cost increase".

③. Model training and optimization
Model performance depends on three key links:

Feature engineering innovation: Construct a "policy sentiment index" to analyze the sentiment value (0-100 points) of the policy texts of the Ministry of Industry and Information Technology and the National Development and Reform Commission through NLP, and quantify the impact of policy severity on costs.

Loss function design: The Huber-Quantile composite loss function is used to ensure the overall prediction accuracy while strengthening the warning capability of extreme market conditions (such as the prediction error is controlled within 5% when the price volatility is >30%).

Real-time calibration mechanism: When the daily fluctuation of raw material prices exceeds 3% or a sudden policy is released, the model is triggered to learn online, and the parameter update is completed within 30 minutes to ensure the timeliness of the prediction.

4. Industry linkage analysis framework
Steel cost forecasting must be placed in the perspective of the global industrial chain for linkage analysis, focusing on the three major transmission mechanisms:

①. Upstream and downstream transmission mechanism
Upstream squeeze effect: In 2025, the cost of iron ore and coking coal will account for 65%. When the price difference between the two widens (such as strong mines and weak coal), the profits of blast furnace steel mills will be compressed, while electric furnace steel mills will benefit relatively.

Downstream demand sensitivity: Automotive sheets are highly tolerant of cold-rolled costs (>6,000 yuan/ton), while construction rebar is highly sensitive. When the cost exceeds 3,600 yuan/ton, demand is significantly suppressed.

Cross-regional arbitrage model: When the price difference between hot-rolled coils in Southeast Asia and China is >50 US dollars, export arbitrage is triggered, and export costs need to be dynamically adjusted.

②. Regional market differentiation modeling
Coastal vs. inland steel mills: Coastal steel mills enjoy the advantages of imported mineral flow (cost is 120 yuan/ton lower), but need to bear the anti-corrosion costs caused by higher humidity; inland steel mills have 8% lower coal procurement costs, but face iron ore transportation premiums.

Climate adaptability adjustment: Northern steel mills increase environmental protection costs by 80 yuan/ton during the winter heating season, and southern steel mills increase the raw material humidity coefficient (+1.5%) during the rainy season, and need to adjust the proportion of ingredients.

③. Policy sensitivity analysis
Construct a quantitative assessment model for policy impact:

text
Environmental protection production restriction cost = benchmark cost × [1 + 0.3×(production restriction ratio) + 0.15×(emission standard coefficient)]
Tariff impact cost = FOB price × (new tariff rate - original tariff rate) × exchange rate adjustment factor
Through the scenario simulation engine, rehearse cost changes under different policy scenarios, such as:

Scenario A: US tariff exemption extension → export cost remains the same

Scenario B: 25% tariff increase → ton steel cost increases by 300 yuan, export volume decreases by 15%

5. Model application and decision support

①. Procurement optimization decision
Based on the cost forecast model, generate a raw material procurement strategy matrix:

Iron ore: When the price drop is predicted to be >5% in the next three months, the inventory cycle will be compressed from 30 days to 20 days, and "small batches and multiple batches" procurement will be adopted.

Alloy: When the price volatility is >10%, option hedging is activated to lock in the highest purchase price, such as buying a call option when the manganese silicon futures premium is 2%6.

Differentiated procurement: Leading enterprises can reduce the cost of coke by 249 yuan/ton through centralized procurement, while lagging enterprises need to optimize channels to reduce price differences.

②. Dynamic pricing model
Embed cost forecasts into product pricing system:

text
Hot-rolled coil benchmark price = (iron ore cost × 0.67 + coke cost × 0.25 + manufacturing cost × 1.1) × (1 + target profit margin)
Market premium rate = 0.4 × automobile production growth rate - 0.2 × real estate investment growth rate + 0.15 × export prosperity index
Realize cost changes to be transmitted to the selling price within 24 hours.

③. Inventory management strategy
Safety inventory: Dynamically adjust according to cost volatility. When the raw material price fluctuates by >15%, increase the coke safety inventory from 15 days to 25 days.

Hedging ratio: When the model predicts the probability of downside risk >60%, increase the futures hedging ratio from 30% to 50%.

④. Investment planning guidance
Optimize capacity layout based on cost forecast: Short process electric furnace: When the scrap steel/iron ore price ratio is <1.8, it is economical and suitable for expansion in the South China region.

Hydrogen metallurgy demonstration project: The initial investment cost increases by 40%, but it has long-term competitiveness when the carbon cost is >100 yuan/ton. It is recommended that the pilot capacity account for 5%-10%.

Conclusion: Build a data-driven cost control system
In 2025, steel cost management has entered the era of algorithm competition. Enterprises need to improve the cost prediction accuracy to more than 95% through multi-source data fusion, dynamic model iteration and industry linkage analysis. The practice of leading enterprises shows that the application of AI cost model can reduce procurement costs by 8%, increase inventory turnover by 2.3 times, and expand gross profit margin by 5 percentage points.

Core value formula:
Cost control benefits = [% increase in prediction accuracy × raw material cost base] + [linked analysis premium × product pricing optimization] - model operation and maintenance investment
Industry evidence: 1 yuan of IT construction investment can get 6.7 yuan of cost savings

With the application of new technologies such as quantum computing and federated learning, steel cost forecasting will evolve towards real-time response and self-optimization. Enterprises that lay out intelligent cost control systems in advance will gain a decisive advantage in the new round of industry reshuffles.