BioPharma Short-Term Forecasting Methods
Robert F. Siegmund, PhD MBA
BioPharma | Insights & Analytics | AI & Data Science | Forecasting & Strategic Planning | Commercial Strategy
DEFINITION
Biopharmaceutical forecasting is the process of generating product forecasts in the biopharmaceutical industry to support decision making. This is important for all parts of the value chain, from making decisions in research and development to making decisions in the supply chain and in the commercialization of pharmaceutical products. It is also important for valuation of biopharmaceutical companies, whether they are publicly traded or privately held.
TYPE OF FORECASTS
The first question is what you are forecasting. I distinguish between production or supply chain forecasts and financial forecasts, there can also be forecast of treated patients with a certain drug. If we start with the production forecast, you will typically forecast grams or kilograms of active pharmaceuticals ingredient (API) or number of vials, syringes, tablets or packs. In a financial forecast you will typically forecast sales in local currency and in a consolidation currency like $ or € for a certain time period. In a financial forecast you can also forecast all the different cost elements: production cost, marketing cost, development costs etc. Derived from that you will calculate a profit and loss account (P&L) and from that you will calculate Net Present Value (NPV) and Return on Investment (ROI). If you forecast treated patients, the forecast will simply contain your number of patients per time frame.
Granularity
Next question is about the granularity of the forecast. There you have three dimensions: market and product, time and time frame, and finally geography. In market and product, it depends if you forecast a whole market or product class, or do you call out the individual products and molecules. Do you forecast only one molecule or do your forecast all the molecules in the product class or all the products in all the product classes in the market. So, for instance in breast cancer do you forecast only HER2+ inhibitors, do you forecast the individual HER2+ inhibitors or do you forecast all the hormone therapies, chemotherapies, immune checkpoint inhibitor therapies, CDKi and so on. So here the main distinction is between the individual product forecast for instance in this example Herceptin (trastuzumab) and the whole market forecast and all variants in between.
Long-Term versus Short-Term
Next time and time frame. You can do a daily, weekly, monthly or yearly forecast and what is the time frame: short term or long term. Here we distinguish between short term, tactical forecasts and long term, strategic forecasts. Supply chain or production forecasts can be both short term and long term, as well as financial forecasts can be both short term or long term. Short term forecasts are typically up to two years or 24 months’ time frame and are typically of daily, weekly or monthly granularity. Long term or strategic forecasts are typically of five to 15 years’ time frame and of yearly granularity
Geography
Regarding geography are different levels of granularity. What we call a level 1 forecast is usually just USA and ex-USA or so-called Rest of World (ROW). Level 2 is broken down by major markets so USA, UK, EU4, Rest of EU, Japan and ROW. And level 3 would be all 30 or 40 countries you need to forecast, forecast individually. Of course, definitions may vary but this is the one I used most.
In-Market versus new product
And finally, an important distinction: in-market versus new product. In-market product means that you are forecasting a brand or product that has been approved, got reimbursement and access and has been launched already. This means you have a lot of empirical data about the performance of the product, and most of all, you have got solid trend data down to a daily level. This makes forecasting so much easier. New product forecast means the product is still in clinical development or at least has not been launched yet, so from a commercial perspective, it’s an unknown entity, you have no market performance data, no trends, no hard empirical evidence.
For this article we will focus on the short-term, in-market forecast.
SHORT-TERM FORECASTING METHODS
A short-term forecast is an in-market forecast where you have ample trends and market performance data about your product available. So, the job is basically to take the historical trend data and write them into the future. The simplest form will be curve fitting, i.e. fitting to a line, i. e. like simply taking a ruler and extending the trend line. You can also fit to a curve that is nonlinear, for instance an exponential equation. For fast growing trends like epidemics exponential equations can be very useful. Spreadsheet calculation programs like Excel have this functionality of fitting data to a linear or non-linear equation built it. Next, I will outline various methods used in short-term forecasting that are being employed by biopharmaceutical companies to forecast revenue, product demand, patient numbers and other outcomes.
1. Linear Regression (Single Variable and Multivariate)
Purpose: Models relationships between dependent variables (like sales) and independent variables (such as time or other factors).
Simple Linear Regression
Focuses on Time as Independent Variable: Models a linear relationship between the dependent variable and time.
where y is the dependent variable, t represents time, β? is the y-intercept, β? is the slope of the line (rate of change over time), and ε is the error term.
Multivariate Linear Regression
Incorporates Multiple Independent Variables: Allows for a comprehensive analysis by including various influencing factors.
Here, x?, x?, ..., x? represent different independent variables, with β?, β?, ..., β? as their respective coefficients.
2. Non-Linear Regression
Captures complex relationships using models like quadratic, exponential, and logarithmic.
Example (Exponential Model): ?
where a and b are constants to be fitted, e is the base of the natural logarithm, and x represents the independent variable. ???
3. Time Series Analysis
Analyzes and forecasts data collected over time.
ARIMA Models (AutoRegressive Integrated Moving Average):???????
领英推荐
Ideal for linear, non-seasonal data. ???????
where L is the lag operator, ?i are parameters of the autoregressive part, θi of the moving average part, and d is the degree of differencing.
Exponential Smoothing: ???????
Focuses on recent observations. ???????
where y^(t+1) is the forecast for the next period, y(t) is the actual value at time t and α is the smoothing constant between 0 and 1.
Seasonal Decomposition: ???????
Breaks down data into seasonal, trend, and residual components. ???
4. Moving Averages
Smoothens short-term fluctuations to reveal underlying long-term trends.
5. Machine Learning Models in Forecasting
Purpose: Utilizes advanced algorithms to learn from data and predict future trends, often handling large datasets and complex patterns better than traditional statistical models
Example:??? Facebook Prophet??????
Designed for forecasting time series data. It accommodates seasonal variations and holidays.?
where g(t) models trend changes, s(t) represents periodic changes (e.g., weekly, yearly), h(t) captures the effect of holidays, and ε? is the error term.???
Other Machine Learning Techniques used in forecast nclude algorithms like Random Forests (of decision trees), Support Vector Machines (SVM), and Neural Networks, which are capable of capturing nonlinear relationships without explicit specification.???????
General Form for a Neural Network:??
where W represents the weights matrix, x is the input feature vector, b is the bias vector, and f is a nonlinear activation function.
CONCLUSION
In the fast-paced world of biopharmaceuticals, the ability to accurately forecast short-term market movements is indispensable. The methodologies discussed—ranging from traditional statistical approaches like linear and non-linear regression, to more contemporary techniques such as machine learning models like Facebook Prophet—provide a robust toolkit for any analyst.
Moreover, it's crucial to remember the importance of incorporating event-based considerations into these forecasts. By integrating both quantitative models and qualitative insights about upcoming events, analysts can significantly enhance the accuracy and relevance of their forecasts. This holistic approach ensures that strategic decisions are both data-driven and contextually informed, offering a competitive edge in the dynamic biopharmaceutical market.
Recommended Literature
If you want to go into more depth on BioPharma Product Forecasting, I recommend the following books.
Tennent, J., & Friend, G. (2011). Guide to Business Modelling (3rd ed.). Economist Books.
Cook, A. G. (2015). Forecasting for the Pharmaceutical Industry: Models for New Product and In-Market Forecasting and How to Use Them (2nd ed.). Gower Publishing.
? Dr. Robert F. Siegmund, Life Code GmbH, Bottmingen, Switzerland, 11.7.2024
Visit us at www.lifecode.ch for the following services