Photo by Markus Winkler on Unsplash

It’s 3 O’clock on a Friday and Bill storms into your office. After firing off a string of expletives that would make a sailor blush, he launches into his usual tirade.

“Why are we out of product XYZ?!”

There are a few answers to that question:

1. The sticky note where you wrote the item, quantity, and customer are lost on your desk.
2. You forgot to make the note.
3. The supplier couldn’t provide the product, but you forgot to tell Bill.
4. You remembered to buy the product, but someone else purchased it before Bill’s customer could.
5. The customer purchased several times more of XYZ than Bill said they would.

Chances are, Bill isn’t happy with whatever answer you gave him. Now, he has to explain to his customer that it will be another 2 to 3 weeks before they get the product he promised you would have when they wanted it.

This is why forecasting matters.

What is inventory forecasting?

Inventory estimation (forecasting) is essentially a process of predicting inventory demand in future time periods. Forecasting can be based on a purchaser’s intuition and anecdotal demand information from sales reps (what is generally referred to as wild guessing) to complex mathematical modeling and everything in between. Each mathematical model, regardless of complexity, makes use of some combination of sales history, trends in sales and demand, and the average lead time to receive new inventory. The aim of forecasting is to balance actual inventory levels with a firm’s customer service requirements. Over-forecasting leads to excess inventory value. Under-forecasting leads to stock-out situations, decreased customer service levels, and increased likelihood that customers will search for the product elsewhere.

It is a common misconception that an increase in the customer service level automatically means a decrease in inventory turnover. As we’ll cover over the next few weeks, it is possible to increase inventory turns and increase the service level at the same time.

Forecast Periods

For most small firms, inventory forecasts will be done on a monthly basis. However, the methods described below can be used in daily, weekly, and monthly forecasting. This is generally dependent on how the inventory management system is designed. A good system will allow the length of time included in a forecast period to vary depending on the velocity of the item (the speed at which the item moves through the firm). For the purposes of replenishment parameters, most mid-market systems support only one level (day, week, month) so the forecast will be aggregated accordingly.

The Guess Methods

According to the U.S. Small Business Administration, a poorly managed inventory system is one of the major reasons small businesses fail, yet an estimated 43 percent of small businesses with fewer than 500 employees don’t have any kind of system to track their goods.

Unfortunately, the various guess methods mentioned previously do not qualify as an inventory management system nor do the generally provide accurate forecasting.

Types of Forecasting Models

Forecasting methods come in two general types:

• Stationary Time Series
• Non-Stationary Time Series

A stationary time series is one where the statistical properties of the process generating the series don’t change over time. This makes them easier to analyze than a non-stationary time series which is one where the properties of the process generating the series are also changing over time.

Shay Palachy has written a great comprehensive primer on Stationarity in time series analysis. While a deep-dive into stationarity won’t be required to understand the concepts throughout Take Stock, it is interesting for the nerdy among us.

Some Stationary Time Series Forecasting Models

A quick mention: there will be an entire article on Take Stock dedicated to each of these models. I will update the article with links to the articles as I upload them.

• Naive Method (NM) This method takes the actual usage from the previous period as the forecast for the upcoming period.
• Moving Average (MA) This method takes the moving average demand from a predetermined number of periods (based on velocity and seasonality) and uses that average as the forecast.
• Exponential Moving Average (EMA) A method similar to the Moving Average that also includes the number of days to forecast as a parameter.
• Non-Zero Periods Moving Average (NZPMA) This method uses the same moving average approach as the Moving Average method but excludes periods with zero usage to calculate a more realistic average for slow-moving and sporadic items.
• Weighted Moving Average (WMA) This method applies weighting factors to prior periods and can be applied to both the MA and NZPMA methods described above. This is like the single exponential smoothing method described below but uses the number of periods based on velocity and seasonality.
• Linear Regression Analysis (LRA) A simple linear regression of the form y=mx+b where the equation is calculated based on historical information. The forecast is then developed based on the time period, x, being forecast. The constants m and b can be recalculated on demand creating a kind of moving average.

What follows is a short list of more advanced forecasting methods that will be discussed on Take Stock.

• Single Exponential Smoothing (SES) A method that utilizes a single smoothing constant to vary the impact that the previous forecast has on the new forecast.
• Syntetos-Boylan Approximation (SBA) This is an approximately unbiased approach to single exponential smoothing. This method helps addresses some of the inherent issues with SES. The magnitude of the error depends on the smoothing constant value being used in the SES model.
• Double Exponential Smoothing (DES) This method makes use of two smoothing constants and can both detect and handle trends in the time series.
• Triple Exponential Smoothing (TES) This is basically the same things as DES but adjusted for seasonality.
• Auto-Regressive Integrated Moving Average (ARIMA) This isn’t a single forecasting method, but a class of standard temporal structures that can be combined to form interesting and complex forecasting methodologies for non-seasonal and seasonal items. The ARIMA for seasonal items is generally abbreviated as sARIMA.

Error in Forecasting Methods

All forecasting methods, no matter how good, will have errors in their predictions. Measuring the error in the preceding models is the only way to determine how accurate each model is.

Error methodologies will be discussed in great detail in the context of each forecast method described above.