Polish Steel Production Under Conditions of Decarbonization—Steel Volume Forecasts Using Time Series and Multiple Linear Regression

Abstract

This paper will discuss the dynamics of steel production in Poland in light of the forecasts of tendencies under conditions of decarbonization. The research presented will be an attempt, using data from 2006 to 2023, to create econometric models and forecast production volumes until 2028, along with influencing factors. The obtained models were compared by calculating their error metrics. Based on the conducted econometric models, the critical determinants of the decarbonization of the industry have been established. Forecasts of the volume of steel production in Poland are downward in the face of the increasingly clear emphasis on strategic investments in low-emission technologies. This paper consists of two research parts. The first concerns the forecasting of steel production volume, and the second concerns the modeling of the steel production process, taking into account the key determinants of technological processes (EAF and BOF). Forecasts were calculated for each econometric model. This analysis is a contribution to a broader discussion on industrial adaptation and sustainable development in the steel sector. The developed models and forecasts can provide decision-makers and industry stakeholders with important information at the stage of the decision-making process concerned with developing a strategy for the decarbonization of steelmaking processes. In Poland, two technologies of steel production are used: BOF and EAF. In accordance with the assumptions of deep decarbonization, BF-BOF technology must be replaced by DRI-EAF technology.

1. Introduction

Industries around the world, over the last few years, have struggled due to severe global crises. The global steel industry has faced significant downturns over the past decade, largely influenced by economic instability, shifting demand, and policy-driven decarbonization efforts. Statistical data from the World Steel Association confirmed that steel production in the European Union fell from 168 million tons in 2010 to just 126 million tons in 2023, marking a decline of nearly 25% over 13 years [1]. This downward trend is not isolated to Europe; China, the world’s largest steel producer, also recorded a slowdown in steel output growth as domestic demand weakened due to economic restructuring and a declining real estate sector. The OECD Global Forum on Steel Excess Capacity [2] has also highlighted overcapacity as a pressing issue, with global steel production consistently exceeding demand, leading to lower prices and squeezed profit margins for many producers. The result of this has been factory closures, job losses, and production cuts, particularly in regions where energy costs and environmental compliance have increased operational expenses [3,4].

Steelmakers in the EU have been among the hardest hit by these economic pressures, largely due to the impact of stringent climate policies that have accelerated the transition away from traditional BF-BOF (Blast Furnace–Basic Oxygen Furnace) production toward lower-emission methods. According to EUROFER (the European Steel Association), the EU has lost more than 34 million tons of annual steel production capacity since 2018, with utilization rates in many plants dropping below 60%—an unsustainable threshold for profitability [5]. In Poland, where steel production was previously above 10 million tons per year, output shrunk to 6.4 million tons in 2023, illustrating the widespread contraction across the region. Additionally, the introduction of the EU Emissions Trading System (ETS) and the Carbon Border Adjustment Mechanism (CBAM) has further increased costs for European steelmakers, compelling them to either pass expenses onto consumers or reduce production, making domestic steel less competitive against imports from less regulated markets like China, India, and Russia [6].

Beyond Europe, the United States and Japan have also experienced contractions in steel output, reflecting a broader global trend. According to the World Steel Association, U.S. steel production declined from 88 million tons in 2019 to 80 million tons in 2023, largely due to reduced demand in the automotive and construction sectors. Similarly, Japan’s crude steel production dropped from 104 million tons in 2018 to 89 million tons in 2023, as domestic industries shifted toward lightweight materials and alternative manufacturing processes. These declines, combined with supply chain disruptions and increasing energy costs, reinforce the reality of a global steel industry downturn that is exacerbated by climate policy measures, evolving trade dynamics, and the rising costs of transitioning toward decarbonized steel production [1].

The Paris Agreement was ratified by at least 55 countries responsible for at least 55% of global greenhouse gas emissions, including all EU countries, on 4 November 2016 [7].

According to the industry policy (Paris Agreement and EU Climate Policies [7]; Green Steel for Europe Initiative [8,9]; EU Emissions Trading System [10]), radical changes will be made to steelmaking, where the existing BF-BOF technology (Blast Furnace–Basic Oxygen Furnace) will be replaced by DRI-EAF technology (Direct Reduced Iron (DRI) in the Electric Arc Furnace) [11,12,13,14,15,16]. Even if green hydrogen is often cited as a prime solution to the problem of decarbonizing steelmaking [17,18,19], it is not the only potential pathway towards reducing the sector’s carbon intensity. Low-carbon alternatives to traditional fossil fuel-based technology exist for the steel sector, and there is more than one available approach beyond using solely green hydrogen [20,21]. Biogenic black carbon from renewable biomass is a potential substitute for fossil-based coke in steelmaking, offering considerable CO₂ emissions savings while maintaining production efficiency. Blue hydrogen produced from natural gas through carbon capture and storage (CCS) is a stopgap solution that still achieves emissions savings as green hydrogen infrastructure becomes available [22]. The viability of these technologies is geographically subject to energy policies, cost competitiveness, and industrial readiness, and thus a diversified approach to decarbonization is essential to ensure the sustainable transformation of the steel industry [23,24].

Now, around 70% of total global steel production relies directly on the input of coal via the BF-BOF route [25]. In the EU, about 60% of steel is produced in BF-BOF processes [26]. In Poland, steel mills use both EAF and BF-BOF technologies. In Poland, 3.3 million tons of electric steel (51%) and 3.1 million tons of converter steel (49%) were produced in 2023 [27]. Now, coal is the most important fuel source in iron and steel production (75%) [28]. Around 1 billion tons of metallurgical coal are used in global steel production, accounting for around 15% of total coal consumption worldwide [1]. This situation is not good for deep decarbonization efforts in global climate policy.

This research was conducted in Poland and focused on the volume of steel production. Several years ago, steel mills in Poland produced more than 10 million tons of steel per year, and in 2023 they produced 6.4 million tons. Currently, steel production is evenly split between two technologies: BF-BOF and EAF (each contributing 50% to total output). Time series analysis (steel production data) from 2026 to 2023 shows strong declines in annual steel production in Poland, as shown in this paper.

The paper applied a broad range of statistical and econometric models to develop an analysis of trends in steel production in Poland and a prognosis by unveiling the characteristics of the sector’s dynamics under the conditions of decarbonization.

This paper presents forecasts for steel production volume in Poland based on actual data from 2006 to 2023, with projections extending to 2028. The analysis relies on annual crude steel production figures (in millions of tons) and employs adaptive econometric methods to predict future trends.

Simple and weighted moving averages, exponential smoothing, Brown’s model, and Holt’s linear models formed an integral part of the advanced exponential–autoregressive frameworks that were applied. These methods were chosen to try and capture the various aspects of the production trends, from short-term fluctuations to the longer-term changes in structure. In all cases, the selected models were checked for their reliability by using error measures, thus making the identification of the most accurate forecast methods possible. In addition, econometric models were developed that incorporated key production variables like coke and scrap consumption in order to reflect the influence of technological and material-specific factors on the production of steel. Forecasts were established for each factor model. This modeling approach provides a robust foundation for understanding the interplay of environmental, economic, and technological factors impacting the steel industry in Poland in its transition to a net-zero industry.

We formulated the following goals of this paper:

• To forecast the volume of steel production in Poland up to 2028 using time series models and multiple linear regression, considering the impact of decarbonization policies.
• To identify and analyze the key determinants influencing steel production, including energy consumption, raw material inputs (coke and scrap), and economic factors.
• To compare different forecasting methods and assess their accuracy, providing industry stakeholders with reliable data for decision-making in the transition toward low-emission steel production.

Our Research Questions (RQs) are as follows:

• RQ1: How will the volume of steel production in Poland evolve in the context of decarbonization, based on time series forecasting methods and econometric models?
• RQ2: What are the critical factors influencing the decline in steel production in Poland, and how do different production technologies (BOF and EAF) impact this trend?
• RQ3: How can econometric modeling, incorporating key industry determinants such as energy consumption, raw material inputs, and economic factors, improve the accuracy of steel production forecasts in Poland?

2. Background of Analysis

2.1. Factors Impacting Steel Industry in Recent Years

Steel production is strongly influenced by market factors [7,12] (fluctuations in steel demand), economic factors (the impact of economic cycles), environmental factors [13,14] (radical requirements to reduce greenhouse gas emissions, especially CO₂), political factors (industrial strategies, financial support instruments, directions for R&D) [15,17] and technological factors [19,20] (long technology life cycles, high investment costs, and other technological barriers to entry for new producers). In recent years, the global economy has experienced periods of economic downturn and recession, particularly impacting industrial economies. This downturn has been characterized by sluggish growth, reduced industrial output, and significant challenges in key sectors, including manufacturing, construction, and energy. These challenges have, in turn, profoundly affected the steel industry, which is closely tied to the performance of industrial economies due to its central role in infrastructure, automotive production, and other manufacturing processes [29].

Steel production is one of the largest industrial sources of carbon dioxide emissions, and its reliance on coal and natural gas further complicates efforts to transition to greener energy sources. The high cost of energy has made investments in more sustainable technologies, such as Electric Arc Furnaces and hydrogen-based production methods, less financially viable in the short term, even though such investments are crucial for the long-term sustainability of the industry [30,31,32,33,34]. Steel production is an energy-intensive process, relying heavily on electricity, natural gas, and coal [35,36,37,38]. As energy prices have soared, the cost of production has escalated dramatically [39]. This has squeezed profit margins and, in some cases, forced steel producers to reduce their output or temporarily shut down operations [40]. In regions where energy costs have become prohibitively high, steel manufacturers have had to contend with the possibility of relocating operations to areas with cheaper energy sources, though such a move is costly and complex.

The combined effects of the energy crisis after the pandemic have prompted steel producers to reconsider their long-term strategies [41,42,43,44]. Many companies are now focusing on diversifying their energy sources, investing in renewable energy, and exploring new technologies to reduce carbon emissions. Additionally, there is a greater emphasis on improving supply chain resilience in order to mitigate the impact of future disruptions. Some producers are also looking to expand into new markets or increase their product offerings to reduce their reliance on traditional sectors that may be more vulnerable to economic downturns [45,46]. Some nations have increased domestic steel production or sought to establish new trade agreements with other steel-producing countries. However, these efforts are often constrained by the time and investment required to develop new production capacity or establish new supply chains.

Table 1. Factors impacting steel industry in recent years.

Factors	Description	Consequences for Steel Industry
Rising Energy Prices [39,45,46]	Surge in electricity, gas, and coal costs due to supply-demand imbalances and geopolitical tensions	Increased production costs lead to reduced profit margins, possible production cutbacks, and higher steel prices. Smaller producers are especially vulnerable to financial challenges or market exit.
Environmental Regulations [47,48]	Stricter emissions policies raising costs for traditional energy sources	Compliance requires investment in cleaner technologies, driving up production costs. Non-compliance risks fines or shutdowns, while adaptation to greener energy demands capital.
Geopolitical Tensions [49,50,51]	Instability in energy-producing regions affecting energy prices and availability	Fluctuating fuel costs and unpredictable supplies increase operating expenses. Companies may face unfavorable contract terms and must frequently adjust to market volatility.
Global Demand Surge [52,53,54]	Post-pandemic economic recovery driving up energy demand	Competition for energy resources leads to cost increases, capacity limitations, and, in some cases, operational suspensions due to insufficient energy supplies.
Infrastructure Limitations [55,56]	Aging infrastructure and inadequate investment causing supply issues	Energy outages and supply bottlenecks affect production, requiring costly upgrades or independent energy solutions which add to operational complexity.
Reduced Fossil Fuel Investments [57,58]	Declining fossil fuel investments limit availability of coal, oil, and gas	Higher costs for essential resources risk production slowdowns, quality issues, and increased costs being passed on to consumers.
Energy Policy Shifts [40,41,42,59]	Rapid changes in carbon pricing and renewable subsidies	Policy uncertainty complicates long-term planning, requiring steel companies to quickly adapt to costly compliance requirements or face reputational damage.
Energy Storage Challenges [60,61,62]	Limited storage for balancing supply with demand fluctuations	Dependence on fossil fuels persists due to inadequate storage for renewables, adding to costs and complicating emission reduction efforts.

Source: Authors’ own work on given references.

shows the main factors impacting the steel industry in recent years.

During periods of economic weakness, investment in large-scale infrastructure projects tends to decrease, as governments and private entities become more cautious in their spending. This reduction in construction activity directly affects the demand for steel, which is a critical material for building bridges, roads, buildings, and other infrastructure. Additionally, the automotive industry, another major consumer of steel, typically experiences reduced sales during economic downturns, leading to a corresponding decline in steel demand [63,64,65,66,67].

Recessions in industrial economies, particularly in Europe, North America, and parts of Asia, have further exacerbated the challenges facing the steel industry [68]. These regions are traditionally among the largest consumers of steel, with their industrial sectors being heavily reliant on consistent steel supplies for manufacturing and construction. As these economies faced slower growth or contraction, the steel industry found itself dealing with oversupply and declining prices [43]. The global steel market, already suffering from overcapacity issues, saw intensified competition among producers, leading to price wars and financial strain on steel companies. Also, the steel industry has been impacted by broader economic trends such as the shift towards a more service-oriented economy and the increasing focus on sustainability. As many industrialized nations have gradually transitioned from manufacturing-based economies to ones dominated by services and technology, the relative importance of heavy industries like steel has diminished. This structural change has contributed to a longer-term decline in steel demand, particularly in advanced economies [32,33]. At the same time, growing environmental concerns and stricter regulations on carbon emissions have placed additional pressure on the steel industry, which is one of the most carbon-intensive sectors globally. Companies have had to invest heavily in cleaner technologies and processes, further straining their financial resources during periods of economic downturn [32,33].

In response to these economic challenges, the steel industry has been forced to adapt in several ways. Companies have focused on consolidating operations, cutting costs, and increasing efficiency to survive the downturn. Mergers and acquisitions have become more common as firms seek to strengthen their market position and reduce excess capacity. Additionally, there has been a push towards innovation, with an emphasis on developing new steel products that meet the demands of a changing market, such as lighter, stronger, and more sustainable materials [49,68].

2.2. Steel Production and Decarbonization of Industry

The data on global crude steel production reveal significant shifts in geographical distribution and overall output between 2012 and 2022. In 2022, the world produced 1885 million tons of crude steel, marking a considerable increase from the 1563 million tons produced in 2012. This growth reflects the expanding demand for steel driven by infrastructure development, industrial growth, and urbanization across the globe. China’s dominance in the steel industry is particularly noteworthy. In 2022, China alone accounted for 54.0% of global steel production, a significant rise from its 46.8% share in 2012. This underscores China’s central role as the world’s largest steel producer, driven by its massive industrial base and infrastructure projects. This increase in China’s share is even more striking considering that other regions have either stagnated or declined in their relative contributions [26].

In contrast, the European Union (EU-27) saw its share of global steel production decline from 10.2% in 2012 to 7.2% in 2022. This decline reflects broader trends of deindustrialization in parts of Europe, as well as the shift of industrial activities to Asia. Similarly, North America, which accounted for 7.8% of global production in 2012, saw a slight decrease to 5.9% by 2022, indicating that, while steel production remains significant, it is not growing at the same pace as in other parts of the world [26,27].

Other regions, like Russia and the Commonwealth of Independent States (CIS), Japan, and India, have also seen changes in their steel production shares. Russia and other CIS countries, along with Ukraine, saw their combined share decrease from 7.1% in 2012 to 4.6% in 2022, possibly reflecting economic challenges and geopolitical tensions. Japan’s share also declined from 6.9% in 2012 to 4.7% in 2022, continuing its gradual decrease as the country faces an aging population and slower economic growth. Conversely, India’s share rose from 4.9% to 6.6% over the same period, reflecting its burgeoning industrial sector and increasing demand for steel in construction and manufacturing [26].

The world needs steel to maintain economic growth. Since 1971, the global demand for steel has increased by a factor of three [69]. The BMI analytical agency forecast a modest growth in global steel production, predicting a 1.2% year-on-year increase in 2024 compared to 2023. This slight growth was expected despite the challenges posed by a deteriorating global industrial and economic outlook. The global economic slowdown, particularly in key manufacturing sectors, was anticipated to exert negative pressure on steel production worldwide. However, robust demand in India was projected to drive the modest increase in global steel consumption for the year [70].

On the pricing front, BMI revised its forecast for average global steel prices in 2024 downward, from an earlier prediction of 740 USD per ton to 700 USD per ton. This adjustment reflects the weaker demand anticipated in China, where the economy continues to grapple with significant risks, most notably within the real estate sector. Since the beginning of the year, steel prices have declined across major markets, with a notable 6.9% drop in China.

Looking further ahead, BMI expects the downward trend in steel prices to persist, with global prices averaging 730 USD per ton in 2025, and then declining further to 675 USD per ton by 2026–2027. This long-term price decline is attributed to the slowing growth of steel consumption in China and the rise of protectionist measures in global markets. These factors are expected to lead to increased production in countries affected by these market shifts, thereby weakening the overall market and putting further downward pressure on prices [70].

The report also highlights a significant paradigm shift in the steel industry, with green steel produced via Electric Arc Furnace (EAF) technology gradually displacing steel produced using traditional blast furnace methods. This shift towards more sustainable steel production methods is expected to gain prominence as environmental considerations become increasingly central to the industry [70].

According to the SMS group, the future of steel production, as projected for 2050, is shaped by a combination of increasing global demand, shifts in production methods, and the urgent need to decarbonize the industry. Current trends indicate that global steel consumption will rise significantly, driven by population growth and industrialization, particularly in populous regions like India and Africa. By 2050, the global population is expected to reach nearly 10 billion, which, at current per capita consumption rates, could require an annual steel production of around 2.2 billion tons. However, as steel demand per capita is expected to increase, production could rise as high as 2.75 billion tons per year [71].

The shift towards renewable energy is poised to significantly impact steel demand. Currently, only 1% to 3% of steel is used in energy infrastructure, but this is expected to grow to more than 10% by 2050, driven by the construction of wind turbines, photovoltaic systems, and the electrification of various sectors. One of the critical challenges for the steel industry is reducing CO₂ emissions while meeting growing demand. The traditional blast furnace route, which currently accounts for 71% of global steel production, is highly carbon-intensive, emitting an average of 1.9 tons of CO₂ per ton of steel. Conversely, the Electric Arc Furnace (EAF) route, which produces steel with significantly lower emissions, currently represents 23% of global production. The availability of scrap, essential for EAF steelmaking, is expected to increase substantially, potentially doubling by 2050. However, only a portion of this scrap will be available for EAF use, limiting the potential increase in EAF-based production [71].

By 2050, it is anticipated that EAF steelmaking could account for 35% to 40% of global steel production, equating to approximately 900 million tons per year. However, the availability of high-quality iron ore, necessary for producing certain grades of steel, remains a limiting factor for the further expansion of this route. The direct reduction (DR) method, particularly when combined with green hydrogen, represents a critical pathway for decarbonizing steel production. However, the growth of this technology is constrained by the availability of high-quality iron ore and the global supply of green hydrogen. Direct reduction capacity is expected to increase from 117 million tons today to nearly 500 million tons by 2050, but this expansion will require significant investments in both plant infrastructure and renewable energy sources [71].

Despite these advancements, the traditional blast furnace route will continue to play a significant role in global steel production, accounting for approximately 40% of output in 2050. Efforts to decarbonize this route include technologies like the Blue Blast Furnace and EASyMelt, which aim to reduce CO₂ emissions by utilizing syngas and integrating renewable energy sources. These innovations offer a more flexible approach to ironmaking, potentially bridging the gap between the availability of high-quality iron ore and the growing demand for green steel [71].

The data also highlight the relatively smaller contributions from regions such as Africa, the Middle East, South America, and Australia and New Zealand, which collectively made up around 6% of global production in both 2012 and 2022. These regions, while important, have not seen the same level of growth or expansion in their steel industries as Asia.

Overall, the data underscore a significant shift towards Asia, particularly China and India, in the global steel production landscape, with other regions either maintaining their output levels or experiencing declines in their relative shares. This reflects broader global economic trends, where manufacturing and industrial activities are increasingly concentrated in Asia, driven by rapid economic growth, urbanization, and infrastructure development in the region.

Continuing economic growth, especially in the developing world, has increased emissions. The global steel industry accounts for about 7% of GHG emissions and 11% of global CO₂ emissions. Reducing industrial CO₂ emissions is crucial to achieving deep decarbonization goals, such as reaching goals of net-zero GHG emissions by 2050 (The Paris Agreement, United Nations, 2015). Achieving the 2050 emission reduction target of 80–95% below 1990 levels necessitates extensive and often radical low-carbon innovation [71].

Government standards, protocols, initiatives and policies have an important role in decarbonizing the global steel industry. The new climate policy—deep decarbonization—involves radical changes in steel manufacturing through low-carbon innovation. European policy has realized a “Green Steel” direction. According to the Green Steel for Europe steel production policy, steelmaking must be clean. Europe is a world leader in environmental sustainability innovation. The technologies of the European steel industry are state-of-the-art and meet BAT requirements. Many steelmaking plants are applying effective solutions for clean steel production. The project “Green Steel for Europe” supports the EU towards achieving the 2030 climate and energy targets and the 2050 long-term strategy for a climate-neutral Europe, with effective solutions for clean steelmaking [70].

Achieving the adopted goal of deep decarbonization requires steel mills to use the VUCA strategy (an acronym for the words Volatility, Uncertainty, Complexity, and Ambiguity) [72]. Companies building strategies according to VUCA should have the ability to function in changing conditions, but be guided by the values and vision of their organization; should respond quickly to volatility; should develop the ability to communicate and obtain information from the environment, i.e., understanding as a response to uncertainty; should develop a clear model for the operation of the company through the talents of employee teams and leadership; need a clarity of function, responsibilities, and process maps for their clash with the complexity of reality; should make decisions faster than other players in the market; should prefer agility as a response to the ambiguity of the environment; should take advantage of opportunities whose durability is usually inversely proportional to the time of decision-making; and, above all, should forecast production and set strategies based on scenarios of possible events. Effective business management in a crisis is not just a matter of management strategy, but is also determined by industries’ access to ICT. In order to make accurate decisions, there is a need to efficiently feed massive data to managers. In the Industry 4.0 approach, heavily popularized since 2011, there are a number of good examples of technological solutions (pillars of Industry 4.0) which are opportunities for the development of the steel industry [73]. Today, strong steel groups in the global steel market are making investments that build smart steel manufacturing [74].

3. Materials and Methods

The methodology used in this paper had two stages. Step 1: On the basis of historical data on the steel sector in Poland, an analysis of the time series of steel production volume for the period from 2006 to 2023 was performed. Then, on the basis of the course of the trend, which was not a clearly upward or clearly downward trend because of its strong fluctuations due to the dynamics of industrial markets and the effects of economic crises (the effects of the US crisis after 2008 as well as the effects of the COVID-19 pandemic) [75], forecasting of steel production volume was performed. The forecasts made, using the methods described in Section 3.1, were ordered by scenario, assuming a downward trend with fluctuations, more positive and more negative (according to the research hypothesis). Step 2: After this first step, econometric modeling (linear regression models) was carried out with the forecasts. This time, the models included factors (determinants) of the decline in the volume of steel production according to technological processes: EAF and BOF. The research performed was used to infer the situation of the Polish steel sector in the conditions of energy and climate policy. A diagram of the research methodology is shown in Figure 1.

3.1. Forecasting Steel Production Volume Using Time Series Models

The volume of steel production (crude steel) in Poland (Mt) was used for forecasting. Figure 2 shows the trend based on the analyzed data. There are quite large fluctuations in the trend of steel production in Poland, which are not cyclical and are difficult to predict. All of the major declines in the volume of steel production in Poland were the result of global problems. In 2009, Polish steel mills produced only 7.1 million tons of steel. This situation was caused by the global economic crisis that began with the destruction of financial markets in the US. In 2013, the second wave of the crisis reached Poland. In 2020, the decline in Poland’s steel production was due to the effects of pandemic restrictions on economies. In the analyzed period, the average annual production was 8.7 million tons. Data on steel production in 2024 were used for comparison with the forecast for that year according to Formula (1):

$$x_{t=2024}={\displaystyle \frac{1-y_{t=2024}}{y_{t=2024}^{*}}}*100$$

(1)

where y_t=2024—the value for 2024 (empirical variable); y*_t—the expired (ex post) forecast value for 2024; and x_t=2024—the forecast percentage error as a percentage discrepancy between the actual data and the forecast for a decreasing trend.

Explanation: Actual data on steel production in Poland in 2024 were published by the World Steel Association after analyses (the report on the WorldSteel website was uploaded on 25 February 2025) [76], but these data were included in the estimation of forecast errors for 2024. In addition, the actual data used for the forecasts had to be identical to the data used for the econometric models. Detailed data on CO₂ emissions by technological processes in Poland, or data on energy intensity, as well as other data used in the models, have not yet appeared in reports for 2024. According to information from the Polish Steel Association, which is responsible for compiling data on the steel industry in Poland, the final report for 2024 will be published in June 2025. For these reasons, econometric modeling and forecasting covered the period from 2026 to 2023.

Figure 2. Steel production (million tons) in Poland in 2006-2023. Source: WorldSteel [77].

Figure 2. Steel production (million tons) in Poland in 2006-2023. Source: WorldSteel [77].

The following research hypothesis (RH) was adopted:

RH: Increasing environmental dynamics will result in a decline in steel production in Poland. It is assumed that there will be a deepening of the downward trend in the volume of steel production in Poland or its flattening below the current average annual production of 8.7 Mt for the period 2006–2023. It can be assumed that, with the continuation of the downward trend, steel production in Poland in the next 5 years will oscillate around 7 million tons (up or down).

The adopted hypothesis stems from the economic downturn in Europe, including reduced investments, which has caused a collapse in the steel market. According to reports, steel production in EU countries has declined. In 2023, the 27 EU countries produced 126.3 million tons of steel, significantly lower than in previous years. Specifically, between 2018 and 2022, the average annual steel production was 149.48 million tons. Despite a slight 2.5% increase in 2024 (129.5 million tons), it remains below the historical average [78].

In the analyzed period for Poland from 2006 to 2023, the average annual steel production in the EU was 165.6 million tons [78]. Additionally, the recent decline in steel production was influenced by the energy market crisis [79,80]. Concerns about radical decarbonization policies for steel producers in EU countries could have also contributed to this drop in production.

The European steel industry is on the brink of collapse; EUROFER warns that the EU must act now or risk losing production [81]. In Brussels, on 27 November 2024, McKinsey warned that the European steel industry is at a critical juncture, facing irreversible decline unless the EU and member states take immediate action to secure its future and green transformation [81].

The following key facts highlight the severity of the crisis:

• Global steel overcapacity reached 551 million tons in 2023—four times the EU’s annual production—and continues to grow. The OECD forecasts an additional 157 million tons of high-emission production capacity by 2026 [81].
• EU steel production has fallen by 34 million tons since 2018, dropping to just 126 million tons in 2023. Imports now account for 27% of the EU market, further weakening domestic production [81].
• Capacity utilization in the EU dropped to an unsustainable 60% [81].
• High energy prices: By 2022, energy costs accounted for 10–18% of BOF steel production and 9–16% for EAF steel. After 2022, energy costs rose to 25–40% due to soaring natural gas and electricity prices. In early 2023, energy prices in the EU fell to levels similar to those at the start of Russia’s invasion of Ukraine, but the energy market remains unstable [82,83].
• Nearly 100,000 steel industry jobs have been lost over the past 15 years, with more cuts expected [81]. Although the EU energy crisis and its impact on the steel industry have eased for now, there is a risk of shifting investments from energy-intensive production sectors in the EU to regions with lower energy costs and decent profit margins. Given competitors with low electricity costs, especially in the US, this factor could weaken Europe’s long-term price competitiveness.
• There are also concerns in Europe about an escalation of the war with Russia and the possibility of emergencies affecting energy raw material supplies [83,84,85].

Forecasts of steel production volumes were determined on the basis of the annual steel production volumes in the period from 2006 to 2023 (t = 18). In 2006, Polish steel mills completed the implementation of repair programs aimed at achieving business stability [86]. After Poland joined the EU in 2004, and Polish steel mills had to meet the requirements of viability and adopt rules to operate according to the requirements of the European steel industry. In 2007, the European Commission recognized that Poland’s steel mills met these requirements. In the period under review, from 2006 to 2023, the average annual steel production was 8.726 million tons.

To forecast steel production, various time series models were employed, each offering unique strengths in capturing trends and fluctuations:

• Simple moving average (SMA)—Models 1a and 1b
  This method averages the last k observations to predict the next value. It is suitable for stable data without significant trends or seasonality.

  $${\widehat{y}}_t={\displaystyle \frac{1}{k}}{\sum }_{i=1}^k{y}_{t-i}$$

  (2)
• Weighted moving average (WMA)—Models 2a and 2b
  WMA assigns different weights to past observations, emphasizing more recent data. This model responds better to recent changes in the series.

  $${\widehat{y}}_t={\sum }_{i=1}^k{w}_i{y}_{t-i}, {\sum }_{i=1}^k{w}_i=1$$

  (3)
• Single exponential smoothing (SES)—Model 3
  SES applies a smoothing factor α to weigh recent data more heavily while considering all past data.

  $$G_t=\alpha y_{t-1}+\left(1-\alpha \right)G_{t-1}$$

  (4)
• Exponential–Autoregressive (Exp-AR)—Models 4a and 4b
  This hybrid model combines exponential smoothing with autoregression, capturing both short-term dynamics and long-term trends.

  $${\widehat{y}}_t={\sum }_{i=1}^k{\beta }_i{y}_{t-i}+\alpha {\sum }_{j=1}^l{\delta }_j{\widehat{y}}_{t-j}$$

  (5)
• Holt’s linear model with additive trend—Model 5
  Holt’s method forecasts data with linear trends using separate smoothing for the level and trend components.

  $$L_t=\alpha y_t+\left(1-\alpha \right)\left(L_{t-1}+T_{t-1}\right)$$

  (6)

  $$T_t=\beta \left(L_t-L_{t-1}\right)+\left(1-\beta \right)T_{t-1}$$

  (7)

  $${\widehat{y}}_{t+h}=L_t+h{T}_t$$

  (8)
• Holt’s linear model with multiplicative trend—Model 6
  This method is similar to the additive model, but assumes trend increases/decreases proportionally over time.

  $$L_t=\alpha {\displaystyle \frac{y_t}{T_{t-1}}}+\left(1-\alpha \right)L_{t-1}$$

  (9)

  $$T_t=\beta {\displaystyle \frac{L_t}{L_{t-1}}}+\left(1-\beta \right)T_{t-1}$$

  (10)

  $${\widehat{y}}_{t+h}=L_t{T}_t^h$$

  (11)
• Brown’s double exponential smoothing—Model 7
  Brown’s method is designed for data with trends, using two levels of exponential smoothing.

  $$S_t^{\left(1\right)}=\alpha y_t+\left(1-\alpha \right)S_{t-1}^{\left(1\right)}$$

  (12)

  $$S_t^{\left(2\right)}=\alpha S_t^{\left(1\right)}+\left(1-\alpha \right)S_{t-1}^{\left(2\right)}$$

  (13)

  $${\widehat{y}}_{t+1}=2S_t^{\left(1\right)}-S_t^{\left(2\right)}+{\displaystyle \frac{\alpha }{1-\alpha }}\left(S_t^{\left(1\right)}-S_t^{\left(2\right)}\right)$$

  (14)
• Advanced exponential–autoregressive model—Model 8
  This refined model integrates multiple smoothing and autoregressive components, capturing complex temporal dynamics.

  $${\widehat{y}}_t={\sum }_{i=1}^k{\beta }_i{y}_{t-i}+\alpha {\sum }_{j=1}^l{\delta }_j{\widehat{y}}_{t-j}$$

  (15)

The above trend-forecasting methods were applied in the following models:

• Model 1a: A simple moving average model for a series formed around a constant (average) value for k = 2.
  Model 1b: A simple moving average model for a series formed around a constant (average) value for k = 3.
  Model 2a: A weighted moving average model for a series formed around a constant (average) value for k = 3, and w₁ = 0.20, w₂ = 0.10, and w₃ = 0.70.
  Model 2b: A weighted moving average model for a series formed around a constant (average) value for k = 3, and w₁ = 0.10, w₂ = 0.10, and w₃ = 0.80.
  Model 3: A single exponential smoothing Brown model for the starting point Gt(2006) and alpha = 0.4421 (by using Solver) for RMSE = 1.1037.
  Model 4a: An exponential–autoregressive model for k = 3, l = 2, beta₁ = 0.70, beta₂ = 0.20, beta₃ = 0.10, delta₁ = 0.60, delta₂ = 0.40, and alpha = 0.34471 for RMSE.
  Model 4b: An exponential–autoregressive model for k = 2, l = 2, beta₁ = 0.60, beta₂ = 0.40, delta₁ = 0.80, delta₂ = 0.20, and alpha = 0.5287 for RMSE.
  Model 5: Holt’s linear model with additive trend for the starting point S₁ = y₂ − y₁ and alpha = 0.7782 and beta = 0.1337 for fi (ψ).
  Model 6: Holt’s linear model with multiplicative trend for the starting point S₁ = y₂/y₁ and alpha = 0.7778 and beta = 0.1296 for fi (ψ).
  Model 7: Brown’s model with double exponential smoothing for the linear trend and alpha = 0.3486 for fi (ψ).
  Model 8: An advanced exponential–autoregressive model for k = 3, l = 2, alpha = 0.4472 for RMSE, and beta₁ = 0.40, beta₂ = 0.50, beta₃ = 0.10, delta₁ = 0.60, and delta₂ = 0.40.
  For these models, k is the smoothing constant; w is weight (weights the used in the forecasting models; l is the number of weights; and S is the starting point.

Models 1 to 8 were used in the analysis. The forecast methodology was performed in accordance with the statistical requirements described in publications [87,88,89,90,91,92,93,94,95,96,97,98,99,100,101]. The weights in the Brown models were set according to his methodology, looking for a smaller forecast error [95]. The methods used were among the most widely used in time series forecasting [97,98,99,100,101]. The models are ordered above from the simplest to the most complex (from a simple moving model to an advanced exponential–autoregressive model). The models used were among the most commonly used in forecasting variables with seasonal fluctuations (when the trend is not an easy trend) on the basis of complete time series, especially when the dynamic status quo is not satisfied.

Classifications and some techniques of forecasting methods are discussed in Archer (1994) [98], Uysal and Crompton (1985) [99], and Witt and Witt (1995) [100]. Time series methods offer concepts and techniques that facilitate specification, estimation, and evaluation, often producing more accurate forecasting results than quantitative causal approaches (Chen et al. 2003) [101]. The most important feature of time series methods is that observations made at different time points are not considered statistically independent. Using appropriate time series methods, the benefits of accurate forecasts can be achieved for production strategies at different time intervals (short-term, medium-term, and long-term), such as planning and investing. This paper chose to use medium-term forecasting, from 2024 to 2028. At the stage of estimating the level of acceptability of forecasting methods, the following ex post forecast errors were estimated: Root Mean Square Error (RMSE) [87,88,89] and ψ [90,91,92,93,94].

$$RMSE=\sqrt{{\displaystyle \frac{1}{n-m}}{{\displaystyle \sum }}_{t=m+1}^n{\left(y_t-y_t^{*}\right)}^2},$$

(16)

where RMSE indicates Root Mean Square Error; m is the number of initial periods or moments of time t for which the expired forecast has not been realized or for which this forecast is the result of a start-up mechanism; yt is the value of the time series (t); and y*_t is the ex post forecast value (l < t < n).

$$\psi ={\displaystyle \frac{1}{n-m}}{{\displaystyle \sum }}_{t=m+1}^n{\displaystyle \frac{\left|y_t-y_t^{*}\right|}{y_t}},$$

(17)

where y_t is the value of the time series (empirical variable), y*_t is the expired (ex post) forecast value (l < t < n), n is number of time series (number of elements in time series), and m is the number of initial periods for which the ex post forecasts were realized.

The forecasting errors were used to select the optimal parameters of the forecasting model used. If the model did not meet the expected assumptions, it was replaced by another forecasting model. The selection of an appropriate model was an important problem determining the quality of forecasts for the volume of steelmaking in Poland for the downward trends.

The results of the forecasts (aggregated forecasts) were used to develop a scenario for the steel industry until 2028. The scenarios were divided into baseline and extreme categories after taking into account the course of the trend-averaged forecasts for the decreasing trends (

Table 2. Ex post forecast values in period from 2016 to 2023.

	Model Numbers
Year	1a (k = 2)	1b (k = 3)	2a (k = 3)	2b (k = 3)	3	4a	4b	5 S1 = y2 − y1	6 S1 = y2/y1	7	8
2006	-	-	-	-	-	-	-	-	-	-	-
2007	-	-	-	-	-	-	-	10.631	10.631	-	-
2008	10.319	-	-	-	10.559	-	-	11.253	11.292	10.442	-
2009	10.179	10.122	9.874	9.846	9.792	9.435	9.685	10.53	10.553	10.02	9.435
2010	8.428	9.162	8.089	7.739	7.465	7.692	7.352	7.993	7.993	7.993	7.783
2011	7.561	8.283	8.254	8.08	8.081	8.128	8.221	8.103	8.098	7.631	8.346
2012	8.385	7.966	8.369	8.533	8.735	8.807	8.774	8.807	8.813	8.067	8.85
2013	8.567	8.376	8.327	8.363	8.383	8.497	8.356	8.591	8.598	8.047	8.458
2014	8.154	8.362	8.156	8.074	8.012	8.082	8.003	8.159	8.164	7.792	8.116
2015	8.254	8.289	8.458	8.478	8.522	8.471	8.577	8.578	8.586	8.127	8.551
2016	8.686	8.441	8.615	8.701	8.763	8.76	8.766	8.894	8.905	8.499	8.746
2017	9.006	8.856	9.032	9.096	9.126	9.035	9.149	9.295	9.312	8.963	9.037
2018	9.764	9.447	9.913	10.065	10.159	9.925	10.221	10.372	10.416	9.978	9.95
2019	10.244	9.895	9.983	10.078	10.081	10.022	10.031	10.454	10.503	10.331	9.914
2020	9.577	9.828	9.38	9.246	9.089	9.156	9.011	9.418	9.457	9.651	9.075
2021	8.427	9.003	8.43	8.2	8.04	8.2	7.987	8.138	8.18	8.487	8.21
2022	8.155	8.536	8.713	8.689	8.752	8.691	8.845	8.617	8.654	8.543	8.824
2023	7.928	8.004	7.627	7.582	7.557	7.883	7.442	7.544	7.591	7.649	7.782
ψ	0.112	0.125	0.101	0.104	1.105	0.106	0.104	0.108	0.11	0.113	0.085
RSME	1.164	1.238	1.061	1.054	1.104	1.014	1.021	1.19	1.206	1.162	0.819

).

The analysis discarded methods (after checking them) for time series with development trends, because the trend studied did not show a strong development trend, and models with high forecast errors, as well as models that did not differentiate forecasts (compared to the model already made in the study).

3.2. Multiple Linear Regression in Forecasting Steel Production

The forecasts based on time series analysis were compared with forecasts obtained using multiple linear regression. Multiple linear regression is a statistical method used to model the relationship between one dependent variable and two or more independent variables. It allows for the estimation of how multiple factors simultaneously influence the predicted outcome. The Formula (18) of the model is as follows:

$$y={\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+\dots +{\beta }_n{x}_n+ϵ,$$

(18)

where $y$ is the dependent variable; $x_1,x_2,\dots , x_n$ are the independent variables; ${\beta }_0$ is the intercept; ${\beta }_1,{\beta }_2,\dots ,{\beta }_n$ are the coefficients; and $ϵ$ is the error term.

In forecasting steel production using multiple linear regression, the explanatory variables in the model (actual data regarding the impact of production factors on the process) were projected based on linear trend models. The forecasts were determined using the following equation:

$${\widehat{y}}_t={\widehat{\beta }}_0+{\widehat{\beta }}_1{\widehat{x}}_{1t}+{\widehat{\beta }}_2{\widehat{x}}_{2t}+\dots +{\widehat{\beta }}_n{\widehat{x}}_{nt},$$

(19)

where ${\widehat{y}}_t$ is the forecasted value of steel production at time $t$, and ${\widehat{x}}_{1t},{\widehat{x}}_{2t},\dots ,{\widehat{x}}_{nt}$ are the forecasted values of the explanatory variables obtained from the linear trend models.

The accuracy of the forecasts obtained using time series models (see previous section) and multiple linear regression was assessed using MAPE, defined by the following formula:

$$\mathrm{MAPE}={\displaystyle \frac{1}{m}}{{\displaystyle \sum }}_{t=1}^m\left|{\displaystyle \frac{y_t-y_t^{*}}{y_t}}\right|,$$

(20)

where y_t is the actual value of steel production at a given time (empirical variable), y*_t is the expired (ex post) forecast value (l < t < n), and m is the number of forecast periods. MAPE is usually expressed as a percentage. According to Lewis [94], forecasts are highly accurate when they have a MAPE < 10%, good with a MAPE from 10% to 20%, average with a MAPE from 20% to 50%, and not accurate when their MAPE is higher than 50%. The lower values of MAPE, the higher the forecast accuracy.

During the modeling stage, the R computer program was used [102] with the packages car, dplyr, lmtest, and psych [103,104,105,106]. The results of the multifactor analysis were correlation matrices using the Pearson coefficient. In modeling the relationships, multiple linear regression models were first estimated with all explanatory variables included in the analysis, and then models with fewer explanatory variables were applied, using backward stepwise regression for variable selection. We present models with one variable, two variables, and more. Each model was statistically evaluated using tests (F-test for linear regression, t-test for independent variables, Shapiro–Wilk test for normality, Breusch–Pagan test for heteroscedasticity), and the results are presented in tables in Section 4.2.

The key innovation of the developed methodology lies in its integrated approach to forecasting steel production, combining time series models with multiple linear regression to enhance predictive accuracy. Unlike traditional forecasting methods that rely solely on historical data trends, this methodology incorporates critical industry-specific determinants, such as energy intensity, coke and scrap consumption, and investment expenditures. By leveraging econometric modeling alongside advanced statistical techniques, this approach not only captures the underlying production dynamics but also accounts for structural changes driven by decarbonization policies and technological shifts. Additionally, the methodology applies a scenario-based forecasting framework, enabling the differentiation between moderate and extreme downward trends, which is crucial for strategic decision-making in a volatile industrial environment.

Further models were also estimated, taking into account the steel production technologies used in Poland. Since 2002, the Polish steel market has been composed of mills using BF-BOF technology and EAF technology. These models used the factors described to analyze these processes (coke consumption, scrap consumption, energy intensity, investment expenditures and fixed assets, employment and salaries, and others). The data included in

Table 3. Steel production by technological process, factors describing processes, and others *.

	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	2023
EAF, thousand tons	4241	4434	45023	3893	3998	4355	4132	3551	3492	3492	3877	4624	4765	4077	3920	4390	3948	3212
BF-BOF, thousand tons	5766	6198	5225	3236	3995	4424	4226	4399	5067	5320.8	5321	5706	5392	4920	3936	4064	3454	3205
Coke consumption, thousand tons	2550	3057	2521	1561	1825	1878	1733	1898	2540	2280	2193	2191	2236	2066	1638	1739	1688	1624 (F)
Scrap consumption, thousand tons	6178	6407	6182	51199	5439	5970	5619	5036	5090	5432	5518	6605	6746	5809	5502	5967	5307	5592
Energy intensity, GWh	2963.1	2983.6	2828.0	2355.6	2505.1	2701.3	2624.4	2297.5	2270.9	2249.1	2426.2	2819.8	2751.6	2480.9	2213.5	2544.9	2497.7	2418.7
CO2 emissions, million tons	12	10.6	8.2	5	7.2	8.18	8.23	8.48	8.6	7	6.7	7.3	6.8	6.3	5	5.5	5.4	5

F—forecast. * Others: Economic factors such as: investment expenditures, employment, fixed assets, and salaries are from: Statistical Yearbooks, Report: Statistical Yearbook of Industry Poland 2024, Warsow, Statistics Poland, available online: https://stat.gov.pl/en/topics/statistical-yearbooks/statistical-yearbooks/statistical-yearbook-of-industry-poland-2024,5,18.html (access data 10 February 2025) [107].

were divided according to technological processes during the development of the econometric models.

4. Results

4.1. Forecasting of Steel Production in Poland Using Time Series

Table 4. Forecasts from 2024 to 2028 in million tons.

Model No. from Table 2	2024	Xt=2024, where y = 7.113	2025	2026	2027	2028	$\overline{\mathit{y}}\mathit{T}$
8.	6.893	−3.19%	6.231	5.570	4.908	4.247	5.570
7.	6.560	−8.43%	6.175	5.790	5.405	5.020	5.790
5.	6.432	−10.59%	6.188	5.945	5.702	5.459	5.945
6.	6.507	−9.31%	6.333	6.163	5.999	5.838	6.168
3.	6.627	−7.33%	6.715	6.754	6.771	6.779	6.729
2b.	6.757	−5.27%	6.788	6.749	6.754	6.757	6.761
4b.	6.615	−7.53%	6.747	6.798	6.824	6.836	6.764
1a.	6.914	−2.88%	6.671	6.792	6.731	6.762	6.774
1b.	6.914	−2.88%	6.914	6.752	6.86	6.842	6.856
2a.	6.990	−1.76%	7.016	6.895	6.926	6.941	6.954
4a.	6.836	−4.05%	6.990	7.136	7.170	7.192	7.065
$\overline{y}t$	6.731	−5.68%	6.615	6.486	6.368	6.243

Source: Own elaboration.

presents the obtained forecasts from 2024 to 2028. The obtained forecasts are sorted in ascending order based on the average of the forecasts for T = 5 ($\overline{y}T$).

After comparing the forecasts for 2024 with the actual data—7.113 million tons of steel being produced in Poland in 2024—an underestimation of the forecasts was identified, but within the accepted error of up to 10% [94], with the exception of model No. 5 in which the underestimation for 2024 was 10.59%. However, referring to the hypothesis of the research and its arguments, it was reasonable to show a negative scenario for the volume of steel production within the limits of the accepted error of forecasts.

The average forecasts ($\overline{y}t$) for t (t = 2024, 2025, 2026, 2027, 2028) formed a downward trend. The obtained values (last column in Table 4) of the projected steel volume are below the actual average annual steel production in Poland in the period from 2006 to 2023 (8.726 million tons). The average annual projected steel production for Poland is 6.489 million tons, which is below the actual steel production in Poland in the period from 2006 to 2023. Figure 3 shows all the obtained forecast trends for steel production in Poland.

The trend of average forecasts was considered the baseline pessimistic scenario and was used to aggregate the remaining forecasts and determine scenarios for steel production in Poland. At this stage of the study, two forecast segments were obtained: a segment with forecasts above the average forecast trend (green color in Figure 4) and a segment below this trend (red color in Figure 4).

The set of models with strongly pessimistic trends (red color) is smaller (four models) than the set of models with moderately pessimistic scenarios (green color), which contains seven models. In addition, the trends of the moderate scenarios are flattened, and the trends of the strongly pessimistic scenarios are strongly downward.

Table 5. Scenarios for forecasting models.

Scenarios of Steel Production
	High Pessimistic Scenario				Basic Pessimistic Scenario	Low Pessimistic Scenario					Towards Optimistic
	Model No.				Last Line From Table 3	Model No.					Model No.
year	8	7	5	6	$\overline{y}t$	2b.	4b.	1a.	1b.	2a.	4a.
2024	6.893	6.560	6.432	6.507	6.731	6.757	6.615	6.914	6.914	6.990	6.836
2025	6.231	6.175	6.188	6.333	6.615	6.788	6.747	6.671	6.914	7.016	6.990
2026	5.570	5.790	5.945	6.163	6.486	6.749	6.798	6.792	6.752	6.895	7.136
2027	4.908	5.405	5.702	5.999	6.368	6.754	6.824	6.731	6.860	6.926	7.170
2028	4.247	5.020	5.459	5.838	6.243	6.757	6.836	6.762	6.842	6.941	7.192
$\overline{y}T$	5.570	5.790	5.945	6.168	6.489	6.761	6.764	6.774	6.856	6.954	7.065

summarizes the projected models according to the two extreme scenarios. In addition, Venn sets were made for all the forecasts of steel production in Poland for the period from 2024 to 2028 (Figure 5).

The resulting sets were ordered according to the minimum and maximum values of the forecasts in each scenario. The results of this analysis are shown in Figure 6.

4.2. Forecasting of Steel Production in Poland Using Multiple Linear Regression

The first table in this section,

Table 6. Steel production—Pearson correlation analysis findings.

Variable	Steel Production	Investment Expenditure	Coke Consupmtion	Employment	Fixed Assets	Salary	Energy Consumption	Scrap Consumption
Steel production	1.00	0.38	0.80 ***	0.36	−0.32	−0.36	0.73 ***	0.79 ***
Investment expenditure	0.38	1.00	0.53 *	0.73 ***	−0.75 ***	−0.63 **	0.56 *	0.30
Coke consumption	0.80 ***	0.53 *	1.00	0.60 **	−0.47 *	−0.53 *	0.55 *	0.47 *
Employment	0.36	0.73 ***	0.60 **	1.00	−0.92 ***	−0.90 ***	0.52 *	0.14
Fixed assets	−0.32	−0.75 ***	−0.47 *	−0.92 ***	1.00	0.92 ***	−0.59 *	−0.19
Salary	−0.36	−0.63 **	−0.53 *	−0.90 ***	0.92 ***	1.00	−0.48 *	−0.10
Energy consupmtion	0.73 ***	0.56 *	0.55 *	0.52 *	−0.59 *	−0.48 *	1.00	0.84 ***
Scrap consumption	0.79 ***	0.30	0.47 *	0.14	−0.19	−0.10	0.84 ***	1.00

Note: *** p < 0.001; ** p < 0.01; * p < 0.05. Source: Own elaboration.

, presents the values of the Pearson correlations between the variables included in the analysis of the relationships between steel production and coke consumption, scrap consumption, energy intensity, investment expenditures, employment, fixed assets, and salaries. The correlation matrix indicates strong positive correlations between steel production and several variables, including coke consumption (r = 0.80), energy consumption (r = 0.73), and scrap consumption (r = 0.79), which are significant at high levels. Some variables, such as employment and fixed assets, show low or negative correlation with steel production, suggesting that they may not be appropriate predictors in the linear model. In the first multiple linear regression model, multiple factors influencing the steel production process were used.

Table 7. A model of the impact of factors on the steel production process—the model with all the variables.

Variable	Estimate	Std. Error	t-Value	p-Value
Intercept	7.521 × 10−1	4.895 × 100	0.154	0.88094
Investment Expenditure	−7.876 × 10−5	3.922 × 10−4	−0.201	0.84486
Coke Consumption	1.585 × 10−3	4.944 × 10−4	3.206	0.00939 **
Employment	−7.891 × 10−2	1.404 × 10−1	−0.562	0.58660
Fixed Assets	7.331 × 10−5	1.184 × 10−4	0.619	0.54956
Salary	−4.299 × 10−4	3.206 × 10−4	−1.341	0.20962
Energy Consumption	9.973 × 10−5	1.529 × 10−3	0.065	0.94928
Scrap Consumption	1.236 × 10−6	6.416 × 10−7	1.926	0.08294

Residual standard error: 0.4967 on 10 degrees of freedom. Multiple R-squared: 0.8921; Adjusted R-squared: 0.8165. F-statistic: 11.81 on 7 and 10 DF; p-value: 0.0004076. Note: ** p < 0.01; p < 0.1. Source: Own elaboration.

illustrates a comprehensive econometric model designed to analyze the impact of various factors on steel production in Poland, incorporating all the potential explanatory variables. The model attempts to capture the intricate relationships among these factors, reflecting the multifaceted nature of steel production processes and their dependence on economic, technological, and environmental conditions. In Table 7, we present the characteristics of this model which included all the variables.

In the model with all the variables, only one was found to be statistically significant at the 0.05 significance level. We decided to apply backward stepwise regression. Ultimately, two explanatory variables were included in the model: coke consumption and scrap consumption (

Table 8. A model of the impact of factors on the steel production process—a model with two variables.

Variable	Estimate	Std. Error	t-Value	p-Value
Intercept	−1.342	1.241	−1.081	0.296752
Coke Consumption	0.001562	0.0003062	5.101	0.000130 ***
Scrap Consumption	0.000001189	0.0000002441	4.871	0.000204 ***

Residual standard error: 0.4566 on 15 degrees of freedom. Multiple R-squared: 0.8632; Adjusted R-squared: 0.8449. F-statistic: 47.31 on 2 and 15 DF; p-value: 3.323 × 10⁻⁷. Note: *** p < 0.001. Source: Own elaboration.

). The set of parameters in the multiple linear regression model is significant, and the individual parameters are also statistically significant (except for the intercept). The following assumptions of the model are met: the homoscedasticity of the random component (Breusch–Pagan test: BP = 0.96149, p-value = 0.6183) and the absence of multicollinearity (the variance inflation factor values are less than 10 for both variables, VIF = 1.290). The assumption of normality of the random component is not met (Shapiro–Wilk test: W = 0.80705; p-value = 0.001914). Additionally, based on Cook’s distance, it can be concluded that there are no outliers (each value is less than 1) (Figure 7).

The inclusion of other variables, such as energy usage and investment expenditures, provided additional context, but showed weaker or statistically insignificant relationships with production volumes. This suggests that, while these factors are relevant, their direct influence may be mediated by other mechanisms or masked by short-term fluctuations in the data.

Based on the estimation model with two explanatory variables, forecasts were determined from 2024 to 2028. This required knowledge of the values of the explanatory variables in these years. The forecasted values of the explanatory variables were obtained using a linear trend. The values of these steel production forecasts are presented in

Table 9. Forecasts of steel production based on multiple linear regression model with two variables.

Year	2024	2025	2026	2027	2028
Forecast	8.067	7.996	7.925	7.854	7.783

Source: Own elaboration.

.

These forecasts indicate that steel production will decline from 2025 onwards, but the forecast values are significantly higher than in the case of the forecasts obtained only on the basis of time series analysis (see Table 3).

In the next step, MAPE was determined for all the time series models and the multiple regression model with two explanatory variables (MR), which allowed for the assessment of the accuracy of expired forecasts in percentage terms and comparisons to be made as to which model was the most accurate (Figure 8).

The multiple regression (MR) model with two explanatory variables achieved the lowest MAPE of 3.38, indicating the highest accuracy among all models. In contrast, Model 1b (simple moving average for k = 3) had the highest MAPE of 12.52, suggesting that it performed the worst in terms of predictive accuracy. The exponential smoothing and Holt models (Models 5, 6, and 7) showed relatively stable errors, ranging from 10.51 to 11.68, implying their moderate forecasting performance. The weighted moving average models (2a and 2b) performed slightly better than simple moving averages but were still less accurate than advanced exponential and regression-based approaches. Overall, the results suggest that the multiple regression model outperforms traditional time series smoothing techniques in this particular forecasting context. It is worth mentioning that forecasts calculated the using multiple regression model are highly accurate, and forecasts based on the different methods are good.

4.3. Forecasting of Steel Production in Poland Using Multiple Linear Regression: BF-BOF and EAF Technologies

Considering that the most accurate steel production forecasts (by MAPE) were obtained using multiple linear regression, it was decided that these models would also be used to forecast production using BOF and EAF technology.

Based on their confirmed relationship, a conclusion was drawn about the statistically significant impact of coke consumption with BF technology and scrap consumption with EAF technology on the volume of steel production. Considering the decarbonization policy of the industry, a reduction in steel production via BF-BOF technology will occur. Therefore, in the second stage of modeling, the focus was on the BF-BOF technology. Statistically significant positive relationships (at the 0.05 level) are clearly visible between steel production and energy consumption, scrap consumption, and coke consumption (

Table 10. BF-BOF steel production—Pearson correlation analysis findings.

Variable	BOF Steel Production	BOF Energy Consumption	BOF Scrap Consumption	BOF Investment Expenditure	BOF Coke Consumption
BOF Steel Production	1.00	0.71 **	0.85 ***	0.28	0.90 ***
BOF Energy Consumption	0.71 **	1.00	0.75 ***	0.00	0.61 **
BOF Scrap Consumtion	0.85 ***	0.75 ***	1.00	0.07	0.71 ***
BOF Investment Expenditure	0.28	0.00	0.07	1.00	0.49 *
BOF Coke Consuption	0.90 ***	0.61 **	0.71 ***	0.49 *	1.00

Note: *** p < 0.001; ** p < 0.01; * p < 0.05. Source: Own elaboration.

). There is a statistically insignificant weak positive relationship between steel production and investment expenditure. A statistically significant positive relationship exists among the explanatory variables (except for investment expenditure). Iron ore was not included in the model because it is a fundamental factor for steel production via BF-BOF technology, and iron ore cannot be replaced by another raw material in this technology. This technology, in line with the decarbonization policy, will be replaced by EAF technology, which requires scrap. In the decarbonization policy, coke, the current reducer, will be replaced by hydrogen (DRI). A description of the model of the impact of factors on the steel production process using BF-BOF technology is presented in

Table 11. A model of the impact of factors on the steel production process using BF-BOF technology—a model with four variables.

Variable	Estimate	Std. Error	t-Value	p-Value
Intercept	−4.825 × 102	6.973 × 102	−0.692	0.50108
BOF Energy Consumption	4.901 × 10−1	1.710 × 100	0.287	0.77888
BOF Scrap Consumption	1.654 × 10−3	7.243 × 10−4	2.283	0.03987 *
BOF Investment Expenditure	−2.253 × 10−1	3.302 × 10−1	−0.682	0.50693
BOF Coke Consumption	1.467 × 100	3.570 × 10−1	4.108	0.00123 **

Residual standard error: 313 on 13 degrees of freedom. Multiple R-squared: 0.9082; Adjusted R-squared: 0.88. F-statistic: 32.17 on 4 and 13 DF; p-value: 1.249 × 10⁻⁶. Note: ** p < 0.01; * p < 0.05. Source: Own elaboration.

.

In the model with all the variables (coke consumption, scrap consumption, energy intensity, investment expenditures and fixed assets, employment, salaries, and others), two variables were found to be statistically significant at the 0.05 significance level. We decided to apply backward stepwise regression. Ultimately, two explanatory variables were included in the model: BOF coke consumption and BOF scrap consumption (

Table 12. A model of the impact of factors on the steel production process using BF-BOF technology—a model with two variables.

Variable	Estimate	Std. Error	t-Value	p-Value
Intercept	−3.789 × 102	4.496 × 102	−0.843	0.41256
BOF Scrap Consumption	1.982 × 10−3	5.415 × 10−4	3.661	0.002319 **
BOF Coke Consumption	1.324 × 100	2.530 × 10−1	5.233	0.000101 ***

Residual standard error: 298.8 on 15 degrees of freedom. Multiple R-squared: 0.9035; Adjusted R-squared: 0.8906. F-statistic: 70.22 on 2 and 15 DF; p-value: 2.421 × 10⁻⁸. Note: *** p < 0.001; ** p < 0.01. Source: Own elaboration.

). The set of parameters in the linear regression model is significant, and the individual parameters (except for the intercept) are also statistically significant. The following assumptions of the model are met: the homoscedasticity of the random component (Breusch–Pagan test: BP = 0.18729, p-value = 0.9106), the normality of the random component (Shapiro–Wilk test: W = 0.96419; p-value = 0.684), and the absence of multicollinearity (VIF = 2.058 for both variables). Additionally, based on Cook’s distance, it can be concluded that there are no outliers (each value is less than 1) (Figure 9). It is noteworthy that the high R-squared value for the model (0.9035) indicates a very good fit of the model to the empirical data. It can be stated that the model raises no objections.

The model with two explanatory variables was used to determine the forecasts of steel production using BF-BOF technology (

Table 13. Forecasts of steel production using BF-BOF technology based on multiple linear regression model with two variables.

Year	2024	2025	2026	2027	2028
Forecast	4150.88	4097.44	4044.00	3990.55	3937.11

Source: Own elaboration.

).

The obtained forecast values indicate that steel production using BF-BOF technology will slowly decline in the coming years. The calculated MAPE is 4.61, which means that the forecasts for steel production using BF-BOF technology are highly accurate.

In the last model, the EAF process was analyzed. A statistically significant positive relationship was found between steel production and energy consumption, as well as scrap consumption. There is a visible positive, moderate, but statistically insignificant relationship between steel production and investment expenditure. The explanatory variables are also positively correlated with each other in a statistically significant way, except for the relationship between investment expenditure and scrap consumption (a positive but statistically insignificant relationship) (

Table 14. EAF steel production—Pearson correlation analysis findings.

Variable	EAF Steel Production	EAF Energy Consumption	EAF Scrap Consumption	EAF Investment Expenditure
EAF Steel Production	1.00	0.79 ***	0.87 ***	0.36
EAF Energy Consumption	0.79 ***	1.00	0.85 ***	0.60 **
EAF Scrap Consumtion	0.87 ***	0.85 ***	1.00	0.35
EAF Investment Expenditure	0.36	0.60 **	0.35	1.00

Note: *** p < 0.001; ** p < 0.01. Source: Own elaboration.

). A description of the model of the impact of factors on the steel production process using EAF technology is presented in

Table 15. A model of the impact of factors on the steel production process using EAF technology—a model with three variables.

Variable	Estimate	Std. Error	t-Value	p-Value
Intercept	−2.625 × 102	6.682 × 102	−0.393	0.7003
EAF Energy Consumption	4.000 × 10−1	5.775 × 10−1	0.693	0.4999
EAF Scrap Consumption	7.731 × 10−4	2.845 × 10−4	2.717	0.0167 *
EAF Investment Expenditure	−2.531 × 10−2	3.241 × 10−1	−0.078	0.9389

Residual standard error: 225.2 on 14 degrees of freedom. Multiple R-squared: 0.7712; Adjusted R-squared: 0.7222. F-statistic: 15.73 on 3 and 14 DF; p-value: 9.216 × 10⁻⁵. Note: * p < 0.05. Source: own elaboration.

.

In the model with all the variables, one variable was found to be statistically significant at the 0.05 significance level. We decided to apply backward stepwise regression. Ultimately, one explanatory variable was included in the model: EAF scrap consumption (

Table 16. A model of the impact of factors on the steel production process using EAF technology—the model with one variable.

Variable	Estimate	Std. Error	t-Value	p-Value
Intercept	−3.700 × 102	6.225 × 102	−0.594	0.561
EAF Scrap Consumption	9.631 × 10−4	1.352 × 10−4	7.125	2.41 × 10−6 ***

Residual standard error: 215.6 on 16 degrees of freedom. Multiple R-squared: 0.7603; Adjusted R-squared: 0.7454. F-statistic: 50.76 on 1 and 16 DF; p-value: 2.411 × 10⁻⁶. Note: *** p < 0.001. Source: Own elaboration.

). In this model, the set of parameters in the linear regression model is significant, and the parameter associated with EAF scrap consumption is also statistically significant. The assumption regarding the homoscedasticity of the random component is met (Breusch–Pagan test: BP = 0.1104; p-value = 0.7397), but the assumption regarding the normality of the random component is not met (Shapiro–Wilk test: W = 0.5316; p-value = 1.526 × 10⁻⁶). Additionally, based on Cook’s distance, it can be concluded that there are no outliers (each value is less than 1) (Figure 10).

The model with one explanatory variable was used to determine the forecasts of steel production using EAF technology (

Table 17. Forecasts of steel production using EAF technology based on multiple linear regression model with one variable.

Year	2024	2025	2026	2027	2028
Forecast	3969.60	3961.12	3952.64	3944.15	3935.67

Source: Own elaboration.

).

The calculated forecast values indicate that steel production using EAF technology will very slowly decline in the coming years. It can even be said that steel production using this technology will remain at a constant level. The calculated MAPE is 3.10, which means that these forecasts for steel production using EAF technology are highly accurate.

5. Discussion

The results of this study present a nuanced view of the steel production trends in Poland from 2006 to 2023 and provide critical insights into how the industry may evolve through 2028 amidst ongoing economic and environmental challenges. The analysis of various forecasting models reveals significant insights into the industry’s potential trajectory, emphasizing the complex interplay between external disruptions and production dynamics.

The comparison of different forecasting methods highlights the variability in predictive accuracy and reliability under current conditions. Among the models analyzed, the advanced exponential–autoregressive model demonstrated the highest forecasting accuracy, with the lowest Root Mean Square Error (RMSE) and average relative error (AVR). This suggests that sophisticated models incorporating both historical data and adaptive smoothing techniques are better suited for capturing underlying trends amidst fluctuating conditions. In contrast, simpler models such as the moving average and Holt linear models, while useful, showed less precision, indicating their limitations in predicting production trends during periods of high volatility.

The results underscore the profound impact of the ongoing energy crisis on steel production in Poland. Rising energy prices, driven by geopolitical tensions and supply chain disruptions, have exerted considerable pressure on the steel industry. The forecasted decline in steel production, as indicated by several models, reflects the industry’s struggle to cope with increased operational costs and reduced profitability. This aligns with the broader industry trend observed globally, where escalating energy costs have led to the scaling back of operations and increased prices for steel products.

The forecast scenarios, particularly those derived from the more accurate models, indicate a trend of fluctuating but generally declining steel production in Poland over the next five years. This trend suggests that the Polish steel industry may face ongoing challenges in achieving pre-crisis production levels. The results highlight the need for strategic adjustments, including investment in energy-efficient technologies and the diversification of energy sources. Moreover, companies might need to adapt to the evolving regulatory environment and market conditions to mitigate the impact of rising costs and supply chain uncertainties.

The scenarios presented in this study are inherently pessimistic due to several converging factors that exacerbate the challenges facing the steel industry in Poland. These scenarios reflect a cautious outlook rooted in the complex interplay of economic, environmental, and geopolitical uncertainties that significantly impact steel production:

• The persistence of high energy prices, driven by global supply chain disruptions and geopolitical tensions, creates a challenging environment for steel producers. Energy costs, which constitute a substantial portion of production expenses, are projected to remain volatile and elevated. This situation is compounded by the ongoing transition away from fossil fuels towards renewable energy sources, which, while necessary for long-term sustainability, entails high initial investments and operational disruptions. The steel industry’s reliance on traditional energy sources, such as coal, means that any instability or increase in energy costs directly affects production efficiency and profit margins.
• The strict environmental regulations imposed on the steel industry add another layer of complexity. Compliance with these regulations often requires substantial investments in cleaner technologies and processes. The anticipated costs of adapting to stricter carbon emission standards and other environmental requirements could further strain the financial stability of steel producers. These regulations are designed to curb emissions and promote sustainability, but they also pose significant economic challenges, particularly for industries that are heavily reliant on carbon-intensive processes.
• The pessimistic scenarios are shaped by the historical data trends and the current trajectory of the steel industry. Analysis reveals a consistent pattern of production declines linked to economic downturns, supply chain issues, and regulatory pressures. Given these historical precedents and the ongoing challenges, these pessimistic scenarios serve as a realistic reflection of the potential outcomes if current conditions persist or worsen.

The study presents a comprehensive analysis of the Polish steel industry’s trajectory under the constraints imposed by decarbonization policies, utilizing time series forecasting and multiple linear regression models. These findings align with previous studies on global steel production trends, where the transition to low-emission technologies has resulted in structural changes within the industry [108,109]. Many industrialized nations have experienced a decline in steel production due to stringent environmental regulations, rising energy costs, and increased competition from regions with more favorable economic conditions [110,111].

Steel companies are strongly linked to energy suppliers, and their LCA analysis of products must take into account many aspects of environmental impact as well as customer requirements (steel consumer markets, e.g., automotive) [112,113].

The observed decline in Polish steel production reflects broader European trends, where traditional blast furnace-based production faces challenges in meeting carbon neutrality goals [32,34,66]. A key contribution of this research is its projection of steel output in Poland until 2028, which suggests a continued downward trend, particularly in the BF-BOF (Blast Furnace–Basic Oxygen Furnace) segment. Similar declines have been observed in other European nations, where steelmakers are progressively shifting towards EAF (Electric Arc Furnace) production, driven by policy incentives and the rising costs of carbon-intensive raw materials [114,115]. In other economies, particularly in Asia, where decarbonization efforts are less aggressive, steel production has either remained stable or increased. This divergence highlights the asymmetric impact of climate policies on global steel manufacturing, with European producers experiencing competitive disadvantages compared to their counterparts in regions with lower environmental compliance costs [116].

The research methodology employed in this study—comparing multiple econometric models and assessing their forecasting accuracy—adds robustness to this analysis. Time series models, including exponential smoothing and autoregressive approaches, have been extensively used in industrial forecasting and have previously demonstrated their utility in predicting manufacturing trends. This study confirms the importance of selecting appropriate models based on error metrics, as different forecasting techniques produce varying degrees of accuracy depending on the underlying data characteristics. The preference for multiple linear regression in this study, particularly models incorporating coke and scrap consumption as key determinants, aligns with earlier research emphasizing the role of material inputs in shaping steel production volumes.

An important consideration in this discussion is the role of external economic factors. Historical downturns in steel production have often been linked to broader macroeconomic shifts, including recessions, fluctuations in global demand, and geopolitical instability affecting raw material supply chains [32,65]. The Polish steel sector, as analyzed in the study, appears to be similarly vulnerable to such pressures. The decline in steel output since 2006 has coincided with financial crises, the pandemic-induced slowdown, and disruptions in global supply chains. In comparison to previous industrial downturns, however, the current trajectory is further compounded by policy-driven changes, making the anticipated recovery in steel production uncertain [75,80].

Another crucial aspect is the transition toward green steel. While Poland, like other European nations, aims to adopt cleaner production methods, this transition entails significant investment and operational restructuring [117,118]. Previous studies on decarbonization in heavy industries have highlighted the financial barriers to adopting new technologies such as hydrogen-based direct reduction. This study’s findings suggest that while Polish steel mills are gradually incorporating EAF technology, this process is constrained by economic feasibility and access to low-carbon energy sources. The shift from BF-BOF methods to the DRI-EAF approach requires substantial infrastructure changes, and its pace will likely depend on external support mechanisms, including subsidies, regulatory frameworks, and technological advancements [9,119].

Comparing these findings with past research on steel industry resilience [120,121] suggests that adaptability will be a decisive factor in determining Poland’s long-term steel production capabilities. Historical case studies from other economies undergoing similar transitions indicate that the success of such shifts hinges on a combination of policy support, market conditions, and industrial innovation. If Poland’s steel sector can effectively leverage EU funding programs and integrate digitalization strategies associated with Industry 4.0, it may mitigate some of the adverse effects projected by this study’s econometric models. However, without strategic investments in green energy and resource efficiency, the industry could face further contraction, particularly if carbon pricing continues to rise.

The application of multiple linear regression models in forecasting steel production volume in Poland led to significantly higher forecast values compared to the time series models. This result can be attributed to the models’ ability to incorporate multiple explanatory variables, such as coke and scrap consumption, which directly influence production processes. Unlike univariate time series methods that rely solely on historical trends, regression models account for industry-specific factors, making them particularly useful for capturing structural changes in the steel sector. The observed discrepancy between the two approaches highlights the importance of integrating economic and technological determinants when making long-term production forecasts.

While regression models provide more optimistic projections, it is essential to consider their underlying assumptions and limitations. The accuracy of these forecasts depends heavily on the stability of the relationships between their independent and dependent variables over time. Any significant shifts in policy, energy prices, or raw material availability could alter these relationships, leading to deviations from the predicted values. Furthermore, given the current trend toward decarbonization and the shift from BF-BOF to EAF technology, future forecasts should continuously be refined by integrating real-time data and scenario-based modeling to enhance their reliability for policymakers and industry stakeholders.

Discoveries of the analysis of Polish steelmaking under decarbonization regulations are owed to the context of path dependency theory, which stresses the role played by earlier choices, the infrastructures in place, and institutional heritages in shaping the historical trajectory of an industry [122,123]. This theory explains that industries stick to a predetermined technological or economic path as soon as it has been ascertained, even if there exists an external need for change [124,125,126]. The Polish steel sector’s reliance on old Blast Furnace–Basic Oxygen Furnace (BF-BOF) technology is due to exactly such inertia, since the transition to cleaner Electric Arc Furnace (EAF) and Direct Reduced Iron (DRI) technologies is hindered by major structural and economic barriers.

Research finds that despite growing emphasis on decarbonization, Polish steelmaking continues to be weighed down by old investments in carbon-intensive production. This is in line with path dependency, whereby industries become trapped in certain technological paradigms since they have such enormous sunk costs and find it so difficult to transition to other modes [127,128]. The Polish steel industry, like all the others in Europe, has been habituated to decades of coal-based production, and therefore a sudden transition to green steel is not possible, except via huge investment and policy-making initiatives. The sustainability of these technologies depends on existing supply chains, skilled workforces, and BF-BOF production facilities, all of which are resistant to change.

This study confirms that even though the regulatory environment, in this case EU climate policy, is inducing change, change is slow. This is a demonstration of self-enforcing path dependency processes, whereby early choices bind up future alternatives for expansion [129]. The shift to EAF or steelmaking on a hydrogen foundation is very costly, thus deterring producers from deviating from their traditional manufacturing process. It is additionally limited by government subsidies, energy price uncertainty, and the green hydrogen supply, all of which have impacts on strategic decisions in the long run.

Another finding of this study is the asymmetrical decarbonization of steel producers. Deep-pocket producers with access to technologically advanced capabilities can readily shift to producing decarbonized steel, whereas small producers remain entrenched in established practices due to their financial position. This serves to reinforce the path-dependent nature of industrial growth, where the lead adopters of technologies accrue competitive benefits, whereas others are left behind [124,127,128].

The predictions in this report, that Polish production of steel will keep on falling, also tend towards the conclusion that path dependency may accumulate in the direction of stagnation or decline unless the industries are overhauled in due course. Since other regions are investing in clean technology in preparation, their steelmaking volumes could remain stable or rise, but the Polish steel industry is at the risk of becoming gradually obsolete if it gets stuck in outdated habits. These can only be broken by focused intervention, such as public investment in low-emission infrastructure, the policy-based inducement of technological innovation, and reconfigurations in supply chains to create room for low-carbon production.

The decarbonization study of the results of the Polish steel industry can also be addressed with an institutional theory, whose central focus is on the impact that regulatory regimes, industry practice, and generalized institutional pressures exert on organizational behavior and industrial transformation [130,131,132]. The research illustrates the way institutional constraints, in particular those arising from EU climate policy, are driving structural change in steel production while also creating huge adaptation issues for firms whose methods are based on traditional types of manufacturing.

Among this research’s findings is decreased steel production in Poland, induced largely by exogenous regulatory impulses rather than simply market impulses. Institutional theory assumes that industries are not autonomous but are shaped by the policy, rules, and standards issued by their regulatory bodies [133,134,135,136]. In such instances, the EU’s 2050 target of carbon neutrality, imposition of the Emissions Trading System (ETS), and Carbon Border Adjustment Mechanism (CBAM) have established an institutional environment committed to low-emission technologies. But the transition to EAF or hydrogen-based steelmaking is more than a response to regulatory pressure; it has to be accompanied by institutions in terms of subsidies, financing tools, and technological assistance.

This report also identifies the asymmetrical response of steel producers to these institutional pressures. A few firms have started investing in cleaner technologies as a way of performing mimetic isomorphism, which institutional theorists refer to as doing what successful industry leaders do in a bid to stay legitimate and competitive. Others are cautious, due to costs or a lack of knowledge of long-term regulatory stability. This variation suggests that while the institutional pressures are high, whether conformity occurs or not depends on firms’ access to resources and industrial location.

Another essential element of institutional theory [137,138] found in our evidence is the function of coercive isomorphism, where legality and policy enforcement force firms into conformity with decarbonization. This study points out that increased environmental compliance expenses, carbon pricing policies, and potentially protectionist trade policy measures have left steelmakers with no choice but to adapt or suffer further economic decline. This adaptation, though, is occurring disproportionately in the sector, since the better institutionally connected players—such as the ones associated with EU-hosted innovation programs—are in a better position to adopt the transition.

The article also refers to the fact that Poland’s steel sector is being confronted with a legitimacy crisis due to its excessive dependence on coal-based manufacturing. Institutional theory contends that in order for industries to prosper they must be legitimate to policymakers, investors, and the broader public [134,135,139]. As international steel markets come under mounting pressure to adopt low-carbon production, policymakers will have a tendency to view adherence to institutional standards as a leading gatekeeper to gaining access to finance, competitiveness, or the establishment of new business alliances. Such a crisis of legitimacy will be quite acute as the EU deepens its green industrial policies, therefore making conformity with sustainability targets a prerequisite to market entry.

6. Conclusions

This study provides a comprehensive analysis of the factors affecting steel production in Poland from 2006 to 2023, with a focus on recent developments and future projections.

The production level of steel in Poland will most probably decline over the next couple of years due to the growing emphasis on decarbonization policies and the shift towards low-emission technologies. Based on time series forecasting methods applied to historical production figures from 2006 to 2023, this study predicts a decreasing steel production, with projections indicating a mean annual production of approximately 6.5 million tons by 2028—well below the historical average of 8.7 million tons. Simple and weighted moving averages, exponential smoothing, and Holt’s linear trend models were used to project future production trends. This analysis assumes that Polish steel production will continue to fluctuate with economic crises, rising energy costs, and the slow abandonment of BF-BOF technology, as this is currently accountable for a considerable share of production. The worst-case scenario, according to the most pessimistic models, estimates a potential fall below 5 million tons, depending on the speed and scale of industrial transformation (RQ1).

Econometric modeling also substantiates these findings by determining key determinants of production, including coke and scrap utilization, energy intensity, and economic determinants such as investment expenditure and fixed assets. Multiple linear regression models confirmed a statistically significant relationship between steel production and these variables, with particular reference to the role of the availability of raw materials as the most critical determinant. The shift from carbon-intensive BF-BOF technology to cleaner EAF (Electric Arc Furnace) technology will require massive investment in green energy, hydrogen, and scrap-processing plants. Based on the modeling outcomes, steelmaking via conventional furnace-based technologies will decline, but the expansion of EAF technology can compensate for this decline with sufficient investment and policy support. However, without strategic interventions, Poland’s steel industry may be exposed to structural issues, including a loss of competitiveness and an increased reliance on imported steel to meet domestic demand.

Polish steel production has been significantly driven downward by a combination of economic, environmental, and technological determinants. The most significant driver among them is the European Union’s decarbonization policy, which subjects carbon emissions to strict regulations, thus making the traditional steel production processes more expensive. BF-BOF (Blast Furnace–Basic Oxygen Furnace) technology, which relies heavily on coke and coal, is the most exposed, due to its high CO₂ emissions. A rise in energy costs also contributes to these issues, as steel is an energy-intensive sector. Economic downturns, reduced investment in industrial plants, and volatile global demand have also led to a decline in domestic steel demand. Increasing raw material prices, particularly metallurgical coal and iron ore, and hefty power charges bring additional pressure on production costs. Geopolitics, e.g., supply shocks and the instability of international steel commerce, also play a role in the downtrend, because Poland comes under greater pressures from low-price EU-outsider producers (RQ2).

The switch away from BF-BOF to EAF (Electric Arc Furnace) technology significantly determines the direction of production. Although EAF is a more efficient and cleaner technology of production, it is dependent on the price of scrap metal and low-cost electricity, which remains a problem in Poland. There is, as yet, immature infrastructure for the recycling of steel and for a green energy supply that is capable of supporting large-scale EAF technology. Moreover, the transformation will require substantial capital investment in new plants, employee retraining, and the updating of existing steel plants, which will curtail output temporarily. In the long term, the shift to EAF will stabilize steel production and align it with sustainability goals. Nevertheless, without strong policy backing and financial incentives, Polish steel production risks shrinking further in terms of quantity as the sector struggles to make economic sense in the context of industrial decarbonization.

Econometric modeling significantly enhances accuracy in forecasting Polish steel production when such industry determinants as raw materials consumed, energy consumed, and economic conditions are included. Whereas conventional time series forecasting relies solely on historical trends, econometric models allow for a sophisticated analysis of the cause-and-effect relationship between production quantity and drivers. Our multiple linear regression models, for instance, illustrate how coke and scrap utilization, energy intensity, and investment expenditure strongly influence steel production. Through the application of these variables, econometric models can also account for structural shifts in the industry, e.g., from BF-BOF technology to EAF technology. This allows analysts to perform alternative policy and market simulations and provide more precise projections of future steel production under evolving economic and regulatory conditions (RQ3).

Econometric models offer enhanced forecasting precision by taking into account external shocks and long-term industry trends. The inclusion of macroeconomic variables—such as GDP growth, global steel demand, and the price volatility of energy—in production models allows one to analyze the effects of the entire economic environment on the industry. For example, the rise in electricity and raw material prices in Poland has a measurable effect on steel production costs and output levels, a fact which would probably be lost when using traditional time series methods. Additionally, statistical model validation techniques, such as the calculation of RMSE (Root Mean Square Error) and MAPE (Mean Absolute Percentage Error), allow researchers to estimate model parameters and select the most precise forecasting technique. Integrating quantitative analysis with real-world industry knowledge, econometric modeling provides policymakers with data-driven solutions to adapting to decarbonization policies, maximizing the use of resources, and maintaining the competitiveness of Poland’s steel sector in a changing market.

The evaluation of the forecasting models in this study was supported by both qualitative insights and quantitative error metrics, ensuring the robustness and reliability of their projections. The comparison of different time series models revealed that the multiple linear regression (MR) model with two explanatory variables (coke consumption and scrap consumption) achieved the lowest Mean Absolute Percentage Error (MAPE) of 3.38%, indicating its highest forecasting accuracy among all the models. In contrast, the simple moving average model (SMA) with k = 3 had the highest MAPE of 12.52%, demonstrating a significantly lower predictive reliability. Additionally, the Root Mean Square Error (RMSE) values further validated model selection, with the MR model having an RMSE of 0.46 compared to 1.23 in the less effective models. These findings quantitatively support the claim that incorporating industry-specific determinants such as raw material consumption improves forecast accuracy over conventional statistical smoothing techniques.

Beyond error metrics, the robustness of the regression model was also validated through statistical significance tests and model diagnostics. The adjusted R² value of 0.8449 for the selected MR model indicates that nearly 85% of the variability in steel production is explained by coke and scrap consumption, confirming the model’s strong explanatory power. The Breusch–Pagan test for heteroscedasticity yields a p-value of 0.6183, confirming that the model meets the assumption of homoscedasticity, while the variance inflation factor (VIF) values for both independent variables were below 2, indicating the absence of multicollinearity. These quantitative measures reinforce the reliability of the model’s predictions, suggesting that the projected decline in Polish steel production—particularly via BF-BOF technologies—will likely continue unless significant investments are made in EAF infrastructure and scrap availability.

The novelty of this paper lies in its integrated approach of forecasting steel production in Poland against the backdrop of industrial decarbonization that utilizes both econometric modeling and time series analysis. Unlike other research, which largely takes into account past tendencies or broad industry projections, this research combines advanced statistical forecasting techniques with the intensive scrutiny of key production factors such as energy consumption, raw material availability, and macroeconomic conditions. This study not only highlights future trends in steel production but also examines the impact of switching from BF-BOF to EAF technology, a key aspect of European Union climate policy. Comparing a number of different forecasting models and testing their validity using statistical measures of error (RMSE, MAPE), this study has a stronger predictive model than standard linear forecasts. This methodological integration brings a unique contribution to bridging the gap between statistical forecasting and economic impact analysis in order to forecast the future of Poland’s steel industry dynamically.

The main scientific contribution of this paper is its developing of a data-driven decision-making framework for the forecasting of steel production tailored to the needs of industrial decarbonization. Through econometric modeling, the study identifies and quantifies the most influential determinants of steel production and delivers information that can guide policymakers and business executives in planning. Its findings indicate that, while Poland’s steel production will decline with higher energy prices and environmental policy, the transition to EAF technology is capable of countering some of these effects with proper investments and policy-making. In addition, the study presents a scenario planning approach to forecasting, detailing several likely futures based on optimistic and pessimistic visions. This allows for a more adaptive and responsive planning process for stakeholders in the steel industry. Overall, this research offers an effective analytical template for assessing long-term steel mill sustainability in Poland and contributes to the broader arguments concerning restructuring heavy industries based on international climate commitments.

It furthers scientific discussion by showing how quantitative modeling can be put together with qualitative insights on policy and strategy. The discussion connects the empirical results to actionable recommendations, such as diversified energy sources, investment in low-carbon technologies, and enhanced supply chain resilience. Such a combination of data-driven analysis with strategic foresight really underlines the practical utility of this research and further develops the general understanding of industrial adaptation in light of systemic disruptions.

This study, while comprehensive, is not without its limitations. One limitation is its reliance on historical data and forecasting models that may not fully capture the dynamic nature of the steel industry’s challenges. The forecasting methods used, including moving averages and exponential smoothing models, provide valuable insights but are inherently constrained by their assumptions and the historical data on which they are based. These models may not adequately account for unforeseen future developments or abrupt changes in external factors, such as sudden geopolitical events or rapid technological advancements. The study primarily focuses on macroeconomic and geopolitical factors affecting steel production, potentially overlooking other significant variables such as technological innovations within the industry, shifts in global trade policies, or changes in consumer demand patterns. The impact of these factors on future production volumes could be substantial, yet they are not deeply explored in this research.

The limitation addressed in this study is the inherent uncertainty in forecasting steel production patterns in the face of evolving market conditions, policy reforms, and technological advancements. While time series and multiple linear regression models provide valuable insights, their predictive accuracy is constrained by the availability and reliability of past data and the assumption that past trends will recur in a similar manner. Additionally, the econometric models used in the analysis may not fully capture the dynamic interactions between the economic, environmental, and geopolitical drivers of the steel sector. External shocks such as uncertainty in energy prices, supply chain disturbances, and shifts in regulation introduce uncertainties that are difficult to model within the existing framework. To mitigate these limitations, scenario-based forecasting was utilized, and this involved having multiple trajectories to capture possible deviations from base-case expectations. Subsequent research needs to blend machine learning algorithms with more refined industry-specific data to further enhance prediction accuracy and adaptability within a changing world of industries.