Forecast vs. Probability of Price Change: A Better Way to Make Commodity Buying Decisions

In commodity procurement, a wrong decision can be expensive. Steel, copper, aluminium and other industrial metals are exposed to volatile markets, shifting supply and demand, geopolitical events and changing macroeconomic conditions. For purchasing teams, anticipating price movements can create a significant competitive advantage.

Traditionally, companies have relied on market forecasts from analysts, consultants and third-party data providers. These forecasts typically tell buyers whether prices are expected to rise or fall, and sometimes provide a specific price target.

But there is a fundamental question worth asking: Is a directional forecast really the most useful information for a purchasing decision?

An alternative is to estimate the probability that a metal price will increase or decrease by a specific percentage over a defined period. Instead of asking where the market will go, this approach asks how likely a meaningful price movement is.

The difference may seem subtle, but it has important implications for transparency, accuracy and commercial decision-making.

1. The problem with traditional price forecasts

Traditional forecasts usually attempt to predict a future price or describe its expected direction. An analyst might predict that copper prices will rise over the next three months, or that steel prices will remain stable.

These forecasts can be useful, but they have several limitations.

Forecasts can accumulate errors over time

Consider a forecast that predicts the price of a metal three months from now.

One way to produce this forecast is to predict the price one month ahead, then use that prediction as the starting point for the second month, and finally use the second-month prediction to estimate the third month.

In simplified terms:

• Month 1: Estimate the price one month from today.
• Month 2: Build on the first estimate to predict the price two months from today.
• Month 3: Build on the second estimate to predict the price three months from today.

If each step introduces an error, those errors can propagate through the forecasting chain. A small initial mistake can affect subsequent estimates, and uncertainty can grow as the forecast horizon extends.

Not every forecasting model works this way. Some models estimate each horizon independently, and forecast errors do not always accumulate monotonically. Nevertheless, a forecast that depends on earlier forecasts can compound errors and become increasingly unreliable at longer horizons.

For procurement teams, the practical question is not simply whether a forecast looks convincing today. It is whether the forecast remains reliable over the period in which a purchasing decision must be made.

Qualitative forecasts can create an illusion of expertise

Many market forecasts combine quantitative data with expert judgement. Analysts interpret economic indicators, assess supply disruptions, evaluate geopolitical developments and form an opinion about future prices.

Expert judgement can provide valuable context, particularly when markets experience structural changes or unprecedented events. But confidence and experience do not automatically translate into forecasting accuracy.

Forecasting research has repeatedly demonstrated that simple quantitative methods can outperform elaborate judgement-based forecasts in many settings. The random walk and other simple benchmarks are particularly important because they are difficult to beat consistently when a market is noisy.

A straight-line or no-change forecast is not universally superior, and commodity prices can exhibit genuine trends. The key lesson is that a sophisticated narrative should never be mistaken for evidence of predictive skill.

A compelling explanation of why prices should rise is not the same as a demonstrated ability to predict that rise.

Third-party forecasts often lack comparable historical accuracy data

Another challenge is transparency.

Many third-party providers publish price outlooks, analyst commentary and market scenarios. However, historical forecast accuracy is not always published in a systematic, independently verifiable format.

Without a consistent record of previous predictions, buyers may struggle to answer basic questions:

• How often has the provider been correct?
• How accurate were its forecasts at one, two and three months?
• How did its forecasts perform during volatile markets?
• Did its forecasts outperform a simple benchmark?
• Were historical forecasts recorded as originally published, or revised later?

Without this information, it is difficult to distinguish genuine predictive ability from persuasive commentary.

This does not mean all third-party forecasts lack transparency. Some providers publish rigorous evaluations. The issue is that buyers should not have to rely on a provider's reputation or confidence alone when assessing the value of its predictions.

2. An alternative: Probability of price change

Instead of predicting a future price, a probability-based approach estimates the likelihood of a defined price movement within a specific period.

For example, rather than saying:

“Copper prices are expected to increase over the next three months.”

A probability-based model might say:

“There is a 75% probability that copper prices will increase by at least 5% within the next three months.”

The exact probability would depend on the model and current market conditions; the figure above is illustrative.

This changes the question from What will happen? to How likely is a commercially meaningful event?

For a purchasing manager, that distinction matters. A predicted price increase of 1% may not justify changing a buying strategy. A substantial probability of a 5% or 10% increase, however, could justify bringing forward a purchase, increasing coverage or reviewing supplier agreements.

Likewise, a high probability of a price decline could indicate an opportunity to delay a purchase, reduce inventory commitments or negotiate more aggressively.

The probability does not eliminate uncertainty. It makes uncertainty explicit and measurable.

3. The advantages and disadvantages of each approach

Neither approach is automatically superior in every situation. A price forecast can be valuable for budgeting, long-term planning and understanding a market scenario. A probability model is particularly useful when the decision depends on whether prices might cross a meaningful threshold.

The most important distinction is whether the information helps a buyer make a better decision, and whether that benefit can be measured.

4. Why horizon-specific probabilities matter

A one-month outlook and a three-month outlook are not interchangeable.

The factors that help explain short-term price movements may differ from those that matter over a longer period. Market momentum, inventory conditions, currency movements, industrial demand and supply disruptions can have different predictive relationships at different horizons.

A probability-based system can be designed and evaluated separately for each time horizon. For example, it can estimate the probability of a 2%, 5% or 10% price increase within one month, two months or three months.

This avoids assuming that a longer-term answer must simply be an extension of a shorter-term forecast.

It also makes evaluation more precise. Buyers can assess whether the model's one-month probabilities are reliable, whether its three-month probabilities are reliable, and where it delivers the greatest value.

Importantly, horizon-specific optimization does not guarantee higher accuracy. Each horizon still needs to be tested against historical outcomes and appropriate benchmarks. The advantage is that the methodology makes such evaluation possible rather than relying on an unverified narrative.

5. Transparency changes the conversation

A probability is useful only if it can be tested.

Suppose a model repeatedly assigns a 70% probability to a particular price movement. Over a sufficiently large number of comparable predictions, that event should occur approximately 70% of the time if the probabilities are well calibrated.

This is the principle of probability calibration.

It provides a concrete way to assess performance. Rather than debating whether an analyst's market commentary sounded convincing, buyers can examine whether the stated probabilities correspond to actual outcomes.

A robust evaluation should include:

• Calibration: Do predicted probabilities match observed event frequencies?
• Historical accuracy: How well did the model perform on data it had not seen during training?
• Benchmark comparison: Does it outperform simple reference models?
• Horizon-specific results: How does performance differ between one-month and three-month predictions?
• Economic value: Would acting on the information have improved purchasing outcomes after accounting for costs, constraints and the buying strategy?

Probability forecasts are not immune to model error, changing market conditions or misleading backtests. Historical performance cannot guarantee future results. But when probabilities are recorded and evaluated systematically, their strengths and weaknesses become much easier to identify.

That creates accountability for the model, the provider and the decisions made using its output.

6. From predicting markets to improving purchasing decisions

Ultimately, procurement teams do not buy metals to win a forecasting competition. They buy materials to fulfil customer demand at an acceptable cost and manage the risks associated with future price movements.

This is where probability-based decision support becomes particularly valuable.

A probability of a significant price increase can help a buyer decide whether to:

• Bring forward planned purchases before a potential increase.
• Secure a larger share of future requirements through fixed-price agreements.
• Increase price coverage when the downside risk of waiting is elevated.
• Revisit inventory policies and purchasing schedules.
• Continue buying normally when the evidence does not justify a change.

Similarly, a high probability of a price decrease may indicate an opportunity to wait, reduce forward commitments or negotiate better terms.

The probability itself is not the decision. The decision depends on the cost of buying early versus waiting, the commercial consequences of a price increase or decrease, supplier flexibility, inventory costs and the buyer's risk tolerance.

This is an important distinction: the objective is not to predict every market movement correctly. It is to make better buying decisions when the financial consequences are meaningful.

7. A real-world example: one buying decision paid back 13% of a project's cost

The idea behind this approach, predicting which size of price move is likely instead of an exact price, is illustrated by a steel wholesaler that used a predictive model to guide a purchasing decision.

The wholesaler did not ask the model for an exact steel price. It asked which price moves actually matter to the business, significant rises and drops over the coming months, and built a model that predicts one of four outcomes: heavy rise, rise, drop or heavy drop. According to Andrea Krogdal's public talk at GAIA, as written up by MetalAlert, the first real decision made with the model paid back 13% of the total project cost.

Read the full case study: How a Steel Wholesaler's Predictive Model Paid Back 13% of Its Cost.

The example highlights the distinction between forecasting a market and identifying a purchasing opportunity. Management faced a purchase it considered too risky; the model said the timing was right, and it went ahead. The model was used as decision support, not as a decision-maker: a way to gain confidence in a call the team already suspected was right.

A single case study does not establish that every buyer will achieve the same result. The figures are those presented in the talk, and results depend on the market, the decision, the model's predictive performance and how the information is applied. Nevertheless, it illustrates how even one well-informed buying decision can have a material commercial impact.

For businesses purchasing significant volumes of steel or other industrial metals, the potential value of improving just a few decisions can be substantial.

Conclusion: Less speculation, more measurable decision support

Traditional forecasts have a place in commodity procurement. They can provide context, describe possible scenarios and support budgeting. But a directional prediction or analyst opinion is not enough on its own.

Forecasts can conceal uncertainty, depend on earlier estimates and be difficult to evaluate when historical performance is not disclosed. Expert judgement can add value, but it should be assessed against simple benchmarks rather than accepted on reputation alone.

Probability of price change offers a different approach. It quantifies the likelihood of commercially meaningful price movements, allows separate evaluation across forecast horizons and creates a transparent record of past performance.

For procurement teams, the shift is from asking “Where do analysts think prices are going?” to asking “How likely is a price movement large enough to affect our costs, and how reliable is that estimate?”

That is a more actionable question, because it connects market intelligence directly to purchasing decisions and measurable financial outcomes.

MetalAlert is built around this principle: helping industrial buyers identify when the risk of rising metal costs is elevated and when an opportunity to reduce purchasing costs may exist. Rather than simply adding another market opinion to the consensus, the goal is to give buyers evidence they can use to challenge assumptions, time decisions more intelligently and align purchasing actions with their real objective: controlling costs.

Read next

• Stop asking for the price forecast. Ask for the price probability
• How a Steel Wholesaler's Predictive Model Paid Back 13% of Its Cost
• Methodology: how MetalAlert builds and tests its probabilities