Techniques for Forecasting Demand for Seasonal Agricultural Products

Every mango farmer knows the rush of peak season – prices crash when supply floods the market, and buyers scramble when supply runs short. This mismatch between what is produced and what is actually needed is one of agribusiness’s oldest and costliest problems. Demand forecasting directly addresses this gap. By predicting how much of a seasonal product buyers will want – and when – agribusiness managers can plan procurement, production, and distribution more precisely, reducing waste and improving profitability. The methods used to do this range from straightforward buyer surveys to sophisticated statistical models, each suited to different data conditions and decision contexts.

Why seasonal agri-products need dedicated forecasting 🔗

Seasonal agricultural products – think tomatoes, mangoes, wheat, onions, or watermelons – do not follow stable year-round demand patterns. Their availability is governed by growing cycles, monsoon timing, cold storage capacity, and regional consumption habits. Agricultural commodity prices are volatile, driven by weather conditions, planting area, consumption trends, and policy factors, making sales decisions difficult for producers and creating instability in incomes and consumer purchasing behavior. A forecast that ignores these seasonal rhythms is simply not useful.

What makes this even more complex is that demand for the same crop can vary sharply between months, regions, and years. A significant characteristic of vegetable price series is seasonality, which is among the biggest obstacles for obtaining accurate forecasts. Without a reliable estimate of future demand, agribusinesses tend to either overproduce – leading to spoilage and price crashes – or underproduce, resulting in missed market opportunities. Accurate forecasting, therefore, is not just a planning tool; it is a risk management strategy.

Survey methods: asking buyers directly 🔗

The most direct approach to forecasting demand is simply asking customers what they intend to buy. Survey methods involve collecting information directly from consumers about their purchasing intentions, either through a complete survey or a sample survey. In agribusiness, this could mean interviewing wholesalers, retailers, food processors, or end consumers about their expected purchases in the coming season.

Complete enumeration vs. sample survey 🔗

In the complete enumeration method, every potential buyer in a defined market is approached and their expected purchase quantities are added together to arrive at a total demand estimate. This is thorough but expensive and time-consuming, making it practical only for smaller, well-defined markets. The sample survey method is more commonly used – a representative group of buyers is selected, interviewed, and their responses are extrapolated to estimate total market demand. The formula used considers the number of households in the market, the proportion reporting demand, and their average expected consumption.

Survey methods are particularly valuable for new products or markets where historical data is thin. However, they carry a key limitation: consumer responses do not always translate into actual purchasing behavior. Stated intentions and real purchases can differ due to income changes, price shifts, or the availability of substitutes at the time of purchase.

Opinion poll methods: leveraging field knowledge 🔗

Where direct buyer surveys are difficult to conduct – such as in dispersed rural markets – opinion poll methods rely on the knowledge of people who interact regularly with buyers and sellers.

Collective opinion (sales force method) 🔗

Under the collective opinion method, the responsibility for estimating expected sales is placed on salespeople, who are closest to consumers and understand their requirements and reactions to products. In agribusiness, this means field agents, input dealers, or procurement officers who operate in specific regions compile their demand estimates, which are then aggregated to get an overall picture. The method is inexpensive and quick, but it is subjective – salespeople may unconsciously understate demand if their sales targets are linked to their forecasts.

Expert opinion and the Delphi method 🔗

In the expert opinion method, the firm seeks out individuals with long-standing experience in the relevant field to provide demand estimates. In agriculture, this could include agronomists, commodity market analysts, extension officers, or experienced traders. When multiple experts are consulted, their estimates may be averaged – either equally (simple average) or weighted based on each expert’s credibility.

A more structured version of this is the Delphi method. Under the Delphi approach, expert estimates are returned to each individual for review, and successive rounds of forecasting take place until consensus improves. A coordinator manages the process, sharing anonymized responses between experts so that each can revise their estimate in light of others’ reasoning – without the risk of groupthink or a dominant voice skewing the outcome. This is especially useful for forecasting demand in new markets or for crops being introduced into new regions, where historical data is scarce.

Time series analysis: mining historical data for patterns 🔗

When reliable historical sales or price data is available, statistical methods become far more powerful than opinion-based approaches. Time series analysis is a commonly used univariate forecasting method that models and analyzes agricultural commodity prices based on the regularity of prices over time, extrapolating future values from existing data. The core idea is that past patterns – trends, cycles, and seasonal fluctuations – tend to repeat, and can therefore be used to predict future demand.

Moving averages and exponential smoothing 🔗

Moving averages calculate predicted demand by averaging data over a fixed number of past periods. This smooths out short-term noise and makes underlying trends visible. Exponential smoothing goes a step further – it gives greater weight to more recent observations while still incorporating older data, making forecasts more responsive to recent shifts in demand. Exponential smoothing and related time series models provide information on demand and what can be expected in the near future, and this information is more reliable than a subjective assessment based solely on managerial experience.

ARIMA and SARIMA models 🔗

For more complex seasonal patterns, the ARIMA (Autoregressive Integrated Moving Average) model and its seasonal variant, SARIMA, are widely used. The seasonal ARIMA model, along with ETS (exponential smoothing) and TBATS models, allows for short-term forecasting as well as identification of time series characteristics. SARIMA is particularly well-suited for seasonal agri-products because it explicitly accounts for periodic fluctuations – like the spike in onion demand during dry seasons or the surge in mango purchases during peak harvest months. Research has reported that the SARIMA model outperforms other price forecasting approaches for forecasting onion prices in Mumbai markets.

The advantage of time series models is that they are data-driven and less subject to human bias. Their limitation is that they rely entirely on historical patterns continuing into the future – a flood, a policy change, or a new competitor can make past data a poor guide to the future.

Econometric models: accounting for what drives demand 🔗

Time series models look only at a product’s own historical data. Econometric models go further – they incorporate external variables that are known to influence demand. Econometric forecasting uses the interplay of demand data with information on external elements that can shift demand to create a demand plan, requiring more sophisticated statistical techniques but potentially yielding a more accurate forecast.

In agriculture, these external variables might include consumer income levels, prices of substitute crops, rainfall data, input costs, government procurement prices, or even export-import policies. A regression-based econometric model might, for example, estimate that demand for wheat flour rises by a certain percentage for every 1% fall in household income or rise in rice prices. Many factors must be considered when assessing and forecasting demand, including cropping patterns, commodity prices, yield levels, prices of competing crops, and disposable farm income.

Regression analysis in demand forecasting 🔗

Regression analysis predicts demand by constructing a mathematical model that expresses the relationship between demand and its key influencing variables. Regression analysis predicts prices by constructing a mathematical model between prices and influencing factors, with methods including linear regression, generalized linear regression, and multiple adaptive regression splines. For agribusiness managers, regression models help answer questions like: if rainfall is 20% below normal, what happens to tomato supply and therefore buyer demand for processed tomato products? The model quantifies these relationships systematically.

A key limitation of regression models is that they assume a relatively stable relationship between demand and its drivers over time. If that relationship changes – for instance, because a new cold chain infrastructure reduces spoilage – the model may need to be re-estimated with fresh data.

Choosing the right method for the right situation 🔗

No single method works best in every situation. The choice depends on what data is available, how volatile the market is, and what decisions the forecast is meant to support. For new products or markets where historical data does not exist, qualitative approaches like consumer surveys and expert opinion polls are the logical starting point. For established products with years of sales records and clear seasonal cycles, SARIMA or econometric regression models are likely to deliver better accuracy.

Even when only limited empirical data is available and other factors influencing demand are unknown, simple mathematical methods can still identify time series patterns for products with strong seasonal fluctuations – making them accessible to smaller agribusinesses that lack sophisticated IT systems. On the other hand, deep learning models like Long Short-Term Memory networks and Gated Recurrent Units have shown superior performance in capturing complex temporal patterns in agricultural commodity forecasting, making them increasingly attractive for larger organizations with access to rich datasets and computational resources.

In practice, many agribusiness managers combine methods – using expert opinion to sanity-check statistical outputs, or blending regression models with time series data – to arrive at a forecast that is both numerically grounded and practically informed. Forecasting also ensures raw materials are secured in sync with harvest and pricing cycles, protecting margins and avoiding costly rush orders – a direct operational benefit of getting the method right.

Impact on agribusiness decision-making 🔗

Accurate demand forecasts shape decisions at every stage of the agribusiness chain. At the farm level, they guide what to plant, how much, and when to bring produce to market. At the processing and trading level, they inform how much raw material to procure and what inventory levels to maintain. At the policy level, they help governments and cooperatives anticipate food security risks and regulate supply.

Accurate forecasting of agricultural commodity prices is essential for market planning and policy formulation, especially in agriculture-dependent economies. When forecasts are wrong – or absent – the consequences ripple through the supply chain: farmers face distress prices, processors run short of raw material, and consumers encounter sudden price spikes. Investing in the right forecasting approach is, therefore, not a luxury for agribusinesses but a core operational requirement.

What do you think? Given that most smallholder agribusinesses in developing countries have limited access to historical price data, which forecasting method – survey-based or expert opinion – do you think would be most practical for them to adopt first? And as digital data collection becomes more accessible, how might that change the case for adopting time series or econometric models even at the smallholder level?

References

1. https://www.mdpi.com/2076-3417/10/22/8169
2. https://pmc.ncbi.nlm.nih.gov/articles/PMC9258887/
3. https://www.vedantu.com/commerce/methods-of-demand-forecasting
4. https://www.slideshare.net/nithineconomics/demand-forecasting-49391812
5. https://mbaknol.com/managerial-economics/expert-opinion-method-of-demand-forecasting/
6. https://www.fao.org/4/t4240e/T4240E14.htm
7. https://www.mdpi.com/2077-0472/13/9/1671
8. https://www.mdpi.com/2071-1050/15/9/7399
9. https://www.waru.edu/acquipedia-article/supply-chain-demand-forecasting
10. https://www.fao.org/4/v4450e/V4450E05.htm
11. https://www.nature.com/articles/s41598-025-05103-z
12. https://acctivate.com/food-manufacturing-demand-planning-and-forecasting/