What is time series modeling, though? First, let’s define the time series itself. A time series refers to a sequence of historical data points ordered in time at successive intervals. These intervals can be annually, quarterly, monthly, or daily.
Time series analysis of data studies the impact of time across a specific variable to identify underlying trends and anomalies. Time series modeling can be for a single variable or incorporate a broader range of inputs. With multivariate time series analysis, there are two or more variables.
Time Series Models
Within finance, time series analysis uses are varied. These models are particularly adept at forecasting future metrics in certain situations. For instance, imagine a company that has consistently launched new products annually during Q4, resulting in a corresponding spike in sales. In this case, time series analysis would be appropriate for forecasting because you can assume the data will follow a similar pattern in subsequent years.
On the other hand, you may not want to utilize a time series model if a recent one-time merger or acquisition significantly impacted a business’s bottom line. This event is unlikely to occur again, and the company may not yet know the long-term synergies associated with the new business venture. In this instance, it may be better to use a different type of analysis for financial modeling rather than relying on a time series forecasting method.
Time series modeling can help guide your projections when you aim to incorporate historical data trends. However, not all previous patterns are applicable, so you must use your judgment to determine if the time series forecasting methods are relevant or if the trend is just an outlier.
Time series analysis uses are best when:
- Data is chronological. When your data follows a linear time frame, utilizing time series analysis can be helpful. Time series modeling relies on chronological data, meaning you can incorporate inputs such as quarterly revenue, monthly sales, or daily stock prices into your analysis.
- Historical trends persist. If you believe past patterns will influence future trends, then time series analysis might be useful. For example, if sales have steadily increased over the past five years, time series modeling would consider this trend when forecasting future sales.
- Cyclicality: Time series analysis of data may offer insight when your inputs follow a seasonal or cyclical pattern. These models can detect and account for such recurring fluctuations. For example, you can consider using time series forecasting methods if sales peak during the holidays.
As with any model, you must evaluate the input data set and output from your time series analysis to ensure a relevant pattern. In some cases, the data may show a trend that is not genuine or is no longer irrelevant.
Further research behind anomalies is always critical to improving the accuracy of your model. For example, an e-commerce site may experience increased sales in one particular year. You can research the business’s marketing strategy to determine if this trend will continue. Consider if the company experienced rapidly growing sales due to a viral social media marketing campaign, for instance. In this case, using the latest growth rate would be misleading as it may not continue in subsequent years.
As you decide if time series analysis is relevant for your forecast, consider these external factors to determine if a trend persists. If there is not, it could be due to a change in market conditions that will impact the business going forward. Time series models may not produce the most accurate results when this happens.
Types of time series models
- Autoregressive (AR) model: An autoregressive model forecasts future outputs by factoring in only past values. The AR model can help forecast sales for the next quarter, for example. These sales will be estimated based on historical sales data.
- Moving average (MA) model: Within time series analysis, an MA model forecasts its output based on the average of the past data points within a specific period. This type of forecasting may be useful for projecting expenses, especially if the company’s monthly costs tend to vary. In this case, an MA model could help determine future budgetary needs by averaging these unexpected fluctuations over a certain period.
- Autoregressive Integrated Moving Average (ARIMA) model: The ARIMA model incorporates both the AR and MA time series forecasting methods. By combining each approach, the ARIMA model contains the best of both. It can use historical trends to forecast the future while also smoothing data to account for outlier trends. ARIMA models can aid in predicting revenue growth for a retail store, for example. This happens by considering historical sales and revenue data while accounting for the cyclical nature of the business.
Forecasting with Time Series
If you can define time series and answer the question “What is time series analysis,” then you already know it can be a beneficial forecasting tool. As with all financial models, though, the strength of time series modeling lies in applying key concepts.
In-sample vs out-of-sample forecasting
In financial modeling, you will use data sets to build the model and test the tool’s accuracy. When the same data serves both purposes, it is called in-sample forecasting. Out-of-sample forecasting involves creating the model with one set of data and then testing its predictive accuracy on a different group. Using out-of-sample data can help you validate your model.
Revenue and expense forecasting
With time series modeling, you need to select the appropriate data set. Consider using past sales data spanning a specific time frame to forecast revenue.
For example, you can analyze a retail business’s revenue over the past two years. In your review, you may discover patterns where revenue surges due to enhanced holiday sales or pinpoint drops in profitability during the off-season. This applies to retail businesses with Black Friday sales and those with heightened demand over the Holiday season. It can also be relevant for office and school supply shops as they will likely experience above-average sales at the start of September and a significant slowing towards the end of the school year.
Even sales of sporting goods are impacted by seasonality as many shops will see demand shift according to the time of year and the sports being played. For example, baseball equipment sales will spike in the spring, only to fall significantly at the end of the summer. These are just a few insights that can be valuable in forming revenue projections using time series analysis.
Similarly, historical costs can form the foundation for expense forecasting. For example, you can evaluate past expenses for a manufacturing company to determine its operational expenditures. When there is increased productivity or demand, costs may rise in certain months due to more significant quality control needs or labor overtime expenses. During periods of elevated production, there may also be higher than usual costs associated with machine maintenance or replacement due to breakdowns. You could even see expenditures fluctuate with the price of raw materials. Time series analysis can help forecast these variations, aiding budget allocations and ensuring a smoother production process.
Projecting best and worst-case revenue scenarios
By studying fluctuations in historical data, forecasters can create both optimistic and pessimistic scenarios. Though forecasts typically highlight the most likely outcomes, you should also acknowledge the possibility of extremes.
Best Case Scenarios
To project a best-case revenue scenario, analyze anomalies, such as periods of maximum growth due to favorable market conditions. You can also consider the impact of a successful marketing campaign or shifting consumer preferences. Next, use these variables to create a “best-case” revenue scenario.
Worst Case Scenarios
You can also study how declining growth could impact the model and build a more conservative projection. For example, imagine how sales would change from an economic downturn or supply chain disruptions. Then, you can use this data to project a “worse-case” revenue scenario.
These best and worst-case scenarios are valuable for a company’s management team. They can also aid investment committees that use time series analysis to assess the risks of a potential investment. The committee can evaluate the probability of a successful venture by reviewing these alternative scenarios.
Applying confidence ranges to financial forecasts
Every model comes with a degree of uncertainty because it is impossible to predict the future. You can quantify the model’s accuracy by applying a confidence interval to your forecast.
For example, imagine your time series analysis forecasts revenue of $1 million next year. You can present this information using a 95% confidence interval of ±$50,000. This tells you there is a 95% chance that the business’s future revenue will be between $950,000 and $1.05 million, allowing for a more precise interpretation of results.
Using FORECAST and TREND functions in Excel
For finance professionals relying on Excel for financial modeling, the FORECAST and TREND functions are time-saving tools. Not only can they offer efficiency, but they also minimize the potential for error as the formulas are automated.
FORECAST: The Excel function predicts a future value by considering historical data. You can use the FORECAST function if, for example, you want to predict next month’s sales based on previous months.
TREND: This function returns values that extend beyond a trend line. It can be helpful when you have a pattern that progresses over time and you want to use time series analysis to forecast additional data points.
Best Practices for Time Series Analysis
In financial forecasting, each model is only as effective as its inputs. Over time, you can fine-tune your process to enhance the accuracy of your models.
Techniques for improving time series models
- Data transformations: At times, raw data can display non-linear trends. Applying transformations, such as square root or log, can help stabilize variance in your data. For example, if a company experiences exponential revenue growth, you can account for the non-linear expansion by taking a log transformation.
- Adjust for seasonality: You can calibrate your model to factor in recurring seasonal patterns, improving your forecast’s accuracy. For example, retail sales might surge during the holidays but drop immediately after. The same might be true for the hours employees work during the season. In this case, you can adjust your data to account for heightened demand during this specific quarter.
- External influences: Factors like inflation rates or economic policies can significantly impact your projections if unaccounted for. If you can identify these factors, consider incorporating them into your model to increase the accuracy of its forecasts.
Measuring forecast accuracy
While countless variables will impact your time series analysis, its output should forecast reality as closely as possible. You can test your model’s capability through out-of-sample forecasting. Using data not included in the model’s initial setup allows you to check if it can accurately forecast the data point. If it does not, then you may need to consider modifying the model.
To gauge the effectiveness of your model, you can also use tools such as mean square deviation, root mean square error, median absolute deviation, and mean absolute percentage error. These metrics can help test for accuracy, allowing you to improve the model where necessary.
Tips for presenting forecasts
While you may have completed your time series analysis, your work is unfinished until you can present it effectively. Here are a few ways to better showcase your forecast to management:
- Visuals: A visual representation using graphs and charts can convey trends more effectively than raw numbers. Within Excel, you can consider add-in tools such as Macabacus to help build compelling visualizations.
- Scenario analysis: Instead of presenting a definite forecast, provide a confidence interval with a range of outputs. Presenting multiple scenarios highlights the inherent uncertainties in forecasting and sets realistic expectations.
- Tailor your wording to your audience: If you present your model to an investment banker, you may want to dive deep into the model intricacies, discussing each variable in detail. However, introducing the model’s projections to the subject company’s management should avoid technical jargon and offer a high-level overview instead.
- Actionable next steps: Rather than presenting numbers and charts only, offer actionable insights based on the model. This will aid management in using the projections to enhance the business’s profitability.
Conclusion
As we answered the question, “What is time series modeling used for?” It is evident that the forecasting method can be a powerful tool when implemented correctly. Time series analysis of data uses historical inputs to make future forecasts, which can help with strategic decision-making.
When you completely understand the concepts behind the forecasting tool, you can realize the advantages of time series in finance. To further refine your modeling capabilities and develop more precise time series forecasts, consider applying best practices in model development and validation.
