Data: The CFOs Secret Weapon for Cash Forecasting Accuracy

Can what’s past be a prologue for cash forecasting? When using statistical modeling and machine learning, yes. CFOs and treasurers have more historical data at their disposal than ever before. If used correctly, forecasts can become more accurate and actionable.

Improving Cash Forecasting Accuracy

In the past, businesses would use simple historical models to predict their future cash flow needs. This meant looking at past cash flow patterns and extrapolating them into the future with basic algebraic means (pun fully intended). However, this method is no longer fully feasible in today’s rapidly changing business landscape, as companies need to quickly adapt and leverage large sets of data to make key financial decisions.

Businesses are now turning to advanced statistical modeling and artificial intelligence to forecast their cash flow needs. This approach offers more accuracy and provides businesses with the flexibility to adapt to changes in their environment.

In today’s market, cash flow really is the lifeblood of the business. Fulfilling short-term obligations such as payroll and counterparty settlements has become challenging for some, as liquidity is strained. Leveraging credit and short-term borrowing is increasingly more costly as interest rates continue to grow.

Cash Forecasting Challenges

Regardless of the sophistication level going into a forecast, organizations first need to have clarity on their historical actuals in order to get a firm grip on future cash flows. Historical data can be sizable and cumbersome to manage when using in-house built systems, or spreadsheets. The ability for those solutions to hold massive amounts of a company’s cash data is unrealistic. Typically data managed in these platforms is very summarized and lacks the granular details needed to properly run various algorithms against.

Additionally, today’s companies are constantly evolving and changing. This means that the cash flow needs of the business today will be different from those in the past.

Quantile Regression in Cash Forecasting

Statistical methods, such as quantile regression, give organizations the ability to estimate the effect of „explanatory variables“ (history), on the distribution of cash collections and overall cash balances (forecasts). These explanatory variables might include the currencies of historical cash coming into the company accounts; the dates of collections, including time of the week and month; types of cash activities; and amounts over the given time horizon. We can use quantile regression to estimate the effect of the historical actuals on the 50th percentile (median) of the distribution, the 75th percentile (upper quartile), the 90th percentile (upper decile), etc.

Unlike most forecasting models that rely on simplistic median and mean-level calculations, quantile regression can be used to estimate the effects of historical actuals on the entire distribution of projected cash. This is important because the distribution of projected forecasts may be very different from the mean or median.

For example, using the mean and median cash balances for the company could project tomorrow’s cash levels at 6 million, but the distribution of historical cash balances is very skewed because of the cyclicality of the balances; several days in history could have balances below the mean and median and a small number of days might have higher cash balance buffers above the mean and median towards the end of the prior period. Comparative to other standard methods of forecasting, quantile regression better estimates the effects of variables that are not linearly related to the underlying response variable, making it less sensitive to outliers.

Implementing Quantile Regression

This is often the topic of discussion these days when it comes to forecasting, as there is not a „one-size-fits-all“ approach. In addition, the elements to the overall forecast may be derived using different methods and modes of calculation.

The one drawback of leveraging advanced calculations like quantile regression in the forecasting process is that it can be computationally intensive. It’s reliant on a large amount of data and parameters to be effective in its accuracy of prediction. Large amounts of historical data, calculated across multiple variables (dates, currencies, cash categorizations, geographies, etc.) put this method out of reach of common forecasting tools like spreadsheets. Rarely does an organization have the available human resources residing in the IT and finance departments to organically develop these models in-house.

Organizations should first look at how to access the necessary data to perform the calculations. This may reside in your ERP, on your ledger, within internal data lakes, or in your treasury management system. To be effective, a larger set of historical data is typically required. Simply basing predictions on 30 days of history, especially considering modern-day volatilities impacting business cash flows, will not be sufficient.

Off-the-shelf technologies can provide finance teams not only access to current and historical liquidity data imperative to mission-critical decision-making, but also provide embedded algorithms to reference this rich history to perform advanced calculations like quantile regressions to support planning purposes.

Lastly, work with your finance and treasury team, identifying the explanatory variables to prioritize in the model, while also eliminating specific variables from the models.

Giving Finance an Edge

In order to navigate the current market turbulence, treasury teams should look to modern technologies for better liquidity planning and management, ensuring that obligations are met, and working capital can be optimized.

By including other sources of data and input into the planning process and implementing advanced statistical calculation methods, finance gains an edge in the critical forecasting of short-term liquidity. Overall, quantile regression is a powerful tool that can be used to improve cash forecasting accuracy.