You can use the budget to create a multi-year cash flow forecast by applying the indirect method. This technique involves combining the income statement and balance sheet to derive the cash flow forecast.

It’s useful for creating a long-term (one year +) forecast but not for forecasting short to medium-term cash flow. This is because it doesn’t have the short-term granularity required for effective short-term forecasting, and it is only refreshed a couple of times a year. A short-term forecast, by its very nature, requires a much more frequent refresh of data and assumptions to be effective.

However, the budget, used in a slightly different way, can support short- to medium-term forecasting.

Again, this needs to be analysed on a per-cash flow category basis. But taking the sales/revenue budget for a medium-term period of, say, six weeks to six months and combining it with payment-timing assumptions is a simple way to translate an income view into a cash flow view. This budget modelling exercise is often the best way to create a medium-term cash flow forecast.

The budget can be translated one-to-one into the forecast for some cash flow categories without the need to apply any assumptions or do any data modelling. For example, stable and easily predictable items such as payroll and rent that aren’t subject to credit terms can simply be copied directly into the cash flow forecast. Depending on the granularity of the forecast, they may need to be manipulated slightly to take quarterly or monthly numbers and translate them into a daily or weekly view.

In summary, the budget is useful for forecasting the following:

• Medium term sales and cost of goods items.
• Short, medium and long term fixed items that aren’t subject to credit terms.
  
  

Statistical & Trend Forecasting

Most statistical and trend forecasting methods combine historical data with mathematical models. Together, they can predict data points in the future.

The amount of data required for this forecast is dependent on the variable you are forecasting and the historical data available for this variable. For example, if you are forecasting customer cash collections for the next 13 weeks — capturing an expected seasonal trend — you’ll likely need a number of years’ worth of data for the model to learn from similar historical periods.

Covid-19 has called into question the use of trend forecasting — particularly when a business was impacted in a profoundly positive or negative way. Using a historical data set that covers the period of most volatility in 2020 is unlikely to produce a useful trend forecast — no matter the model used.

However, as the economy normalises and businesses attempt to look through the impact of Covid, the use of historical data to build trend models becomes viable again. This is particularly true for businesses benefitting from shorter-term trends (week-on-week, month-on-month, etc.).

Of course, statistical forecasting is about more than just extrapolating a historical trend. Trend forecasting is one of the techniques most relevant to cash forecasting. Some useful statistical forecasting models include:

Naive

A naive model is the simplest of all forecasting methods as it simply rolls historical data into a future period. It can be an excellent baseline for which assumptions are applied and for comparison instead of more involved methods to test their credentials.

Moving Average

This is a time series technique that takes the average of a historical data set — or periods within the data set — and uses these as the basis for the forecast. The moving average calculation can vary in complexity:

• Simple moving average: Treats all historical data points equally
• Weighted and exponential moving averages: Weighs recent data more heavily within the model

Linear Trend

A linear trend model takes a historical data set and “fits” a line that best represents the behaviour the model represented in the data. This line then becomes the forecast which represents the historical trend.

Other Time Series

A number of other time series methods exist — notably AutoRegressive Integrated Moving Average (ARIMA) and exponential smoothing. Both of these techniques fit trend lines to data sets and use these as the basis for the forecast. But they are used in different ways in different types of data sets.

As with all other forecasting techniques, it is important to understand the nature of the data and the type of forecast being built before selecting a statistical forecasting method. Trend and statistical models should be used when:

• The past behaviour of a cash flow line item is representative of expected future behaviour.
• A meaningful amount of data is available to allow the model to learn from past trends.
  
  

Driver Based Forecasting

Driver-based forecasting ties in very closely with trend and statistical forecasting. In a driver-based model, the relationship between two variables is analysed, and the forecast of one variable is then used to “drive” the
prediction of another variable.

A simple example is using a revenue forecast combined with expected future gross margins to forecast the expected cost of goods expense. Similarly, driver-based forecasting is particularly valuable to industries with exposure to and reliance on a number of key input costs, such as oil or other commodity prices. In this case, a forecast of input volume required combined with an expectation of the commodity price will produce a forecast of the required input cost. The input volume could be driven by the expected demand for the company’s product. Therefore, driver-based forecasting is often multi-layered.

Driver-based forecasting is particularly useful for scenario analysis. This is because changing the input variables (such as volume or price, in this example) will cause a subsequent change in expected input costs. So if oil goes up by X, our costs go up by Y.

A number of key line items in the cash forecast can be drivers for other parts of the forecast. So, while all of your forecast may not be driver-based, some parts of it will be.

Driver-based forecasting is useful when:

• There is an established relationship between two cash flow line items or another non-cash flow input (e.g., the
  price of oil).
• The variables used to drive the forecast can be forecasted with a reasonable degree of accuracy.