Price forecasting of components purchased for electronics equipment is not new, but has become more strategic to many companies as it helps OEMs price their end products in the marketplace.

"Price forecasting is extremely important to us. It is a critical process," says Patrick Scholler, procurement director for Hewlett-Packard, based in Palo Alto, Calif. "Our margins are thin and any variation in the material cost can have a dramatic impact on a given business-unit performance."

It has such an impact because material costs can be about 75–85% of the overall cost of an HP computer. Inaccurate forecasting can have an adverse impact on HP's earnings.

When HP salespeople quote prices for finished systems to customers, they use the materials price forecasts that were developed by purchasing to come up with a bid.

"We have a significant portion of our business quoted with a three-month horizon. In some deals, it could be six months or more," says Scholler.

Accurate price forecasts for components are crucial to the bidding process. "If the price forecasts are too high, it means that our quotes may not be competitive and we will lose the deal. If they are too low, we will probably win the deal, but we may lose our margin and that will have an impact on our earnings," he says.

Scholler says every part number has a forecast and most forecasts are updated monthly. The forecasts go out 12 months. However, prices for some products such as DRAM and liquid crystal displays (LCDs) are changeable and forecasts are updated more often, says Scholler.

"We spend a lot of time on DRAM because it is so volatile," he says. "It's like predicting what will be the price of a stock six months from now." DRAM forecasts are updated every two weeks and sometimes every week.

Price forecasts are market-based, cost-based or a combination of the two.

"Market-based pricing is driven by the balance between supply and demand," says Scholler. "Sometimes the price is cost-based and is more dependent on some raw materials costs."

With cost-based products, the capital expenditures (capex) that suppliers make are factored in because "capex plays an extremely big role in cost," he says. Raw materials and labor costs are also factored in.

"We do cost modeling and try to understand every cost that is contained with that part," he says.

Cisco Systems, based in San Jose, Calif., also does cost modeling and has developed a method to determine costs. It looks at the capital expenditures of semiconductor suppliers and the depreciation in schedules for equipment and plants.

"We also look at the cost of raw materials like silicon, the cost of a wafer and obviously the availability of wafers," says Jody Kail, commodity director for Cisco's global supplier management group.

"As far as semiconductors are concerned, raw materials are about 10% of the overall cost. There is not a lot you can do except watch that silicon price closely," he says.

Cisco also monitors the price of oil. "We look at the price of a barrel of oil and project our productivity strategies based on that," says Kail.

He says any kind of petroleum-based products like photo resist will be affected by the price of oil. "It's also an issue with plastics used in chip packaging," he says.

"Printed circuit boards can be impacted by the price of oil," says Kail. "With boards you have a much higher level of special metal content. (A cost of) $950/ounce for gold has a small effect." The price of copper has more of an impact.

With semiconductor cost, Cisco also gets input from its key suppliers concerning cost. "We are working with all of the major semiconductor suppliers and have a good idea of cost per square inch of silicon and what it will be in the future," he says.

With passive components, raw material costs impact component prices, but not in the short-term, according to Eric Pratt, executive director, pricing and competitive analysis for researcher iSuppli in El Segundo, Calif..

"Raw materials are a good long-term indicator because pushing price increases through to market for raw materials becomes more difficult. It is not a quarter on quarter thing," he says.

Passives suppliers often absorb material price increases for awhile because they want to stay price-competitive in the market and not lose market share.

With market-based price forecasts it is important to forecast future demand for a component. For instance, if HP buyers are forecasting the price of memory chips they will try to determine how many computers are expected to ship. They will also try to determine the expected memory content in systems because it tends to increase every year.

"Will PC makers put more memory into their systems? Will the new operating system from Microsoft require more memory? That's what we ask," says Scholler.

It is also important to determine how much capacity will be added to meet rising demand. HP tries to determine how much capex each semiconductor company will make and if the investments will significantly increase capacity. It also wants to know if suppliers are transitioning from 8-inch to 12-inch wafers.

"We look at those changes for each supplier. We know supplier by supplier what the plans are and we factor that in and include that in our price analysis," says Scholler.

He says that process changes, such as moving to 12-inch wafers or to 65nm process technology from 90nm results in more chips per wafer, but the extra capacity may not immediately kick in.

A process change can increase capacity, but the change may not go smoothly. "There may be a yield or quality problem that may result in shortages and it may take several months to correct the issue," says Scholler. Buyers need to monitor those changes at a supplier and determine how it may affect chip output.

While buyers need to know the process changes and capex plans of suppliers, they must also analyze where the investment and changes are being made.

A supplier could increase its capex for one technology while not spending any money at all on another, says Pratt. A lot of capex is made for new technologies rather than mature technology. A company could invest in digital signal processor technology, but not in more mature products such as standard logic or analog.

A memory supplier may shift capacity and make process technology changes for DRAM, but not be as aggressive with capex for flash memory.

How well a buyer can determine both demand and capacity is critical to accurate price forecasting. It is rare that price forecasts are 100% exact, especially for volatile parts such as memory and LCDs. The accuracy of a forecast also drops off the farther out it goes.

Pratt says that with DRAM or flash, a forecast within 5–8%, 90 days out is fairly accurate because memory prices are volatile.

"With more stable commodities like analog semiconductors and connectors and some resistors you can be very accurate one year out and be within 5%", Pratt says.

Sometimes a buyer's price forecast will be too low or too high consistently and this may indicate "price bias," according to Scholler.

"Bias is when you systematically forecast too high or too low," says Scholler. Bias can mean that too much weight is given to certain factors that determine demand, capacity or the supplier's cost to build a part.

In some cases it could just be a buyer forecasting a part too high so he can later purchase the part for less than the forecast.