The covid coefficient: Adapting your business forecasting techniques for a pandemic

By
on
Dec 15, 2020

In these times of unprecedented uncertainty, how do you plan ahead? Here are a few options for adapting your business forecasting techniques during the Covid-19 pandemic.

In these times of unprecedented uncertainty, how do you plan ahead? It’s virtually impossible to know which way the economy will turn next, but your business still needs budgets and forecasts in order to function.

The gini team spoke to Kevin Pereira, university lecturer on Artificial Intelligence and Big Data, and Managing Director of AI consulting firm Blu Ltd., about how to adapt business forecasting techniques during the pandemic.

We’ve never seen anything like this before

Usually, when an economic shock happens, you’d estimate the impact by looking at what happened last time, or in similar industries or regions. But there’s nothing usual about Covid-19. 

With no comparable past, business forecasting techniques perfected over the years are suddenly no longer applicable. No machine learning model has been trained to account for something like this. So what do you do? 

“First, you need to figure out whether you think the economic impact of Covid-19 will be short-term or long-term for your industry,” says Kevin. “That’ll help you figure out which business forecasting technique makes the most sense for you.”

The way you adapt your forecasting technique depends heavily on your business, industry and use case. In this article, we’ll focus on regression analysis techniques for time series data, as that is the most common method used for business forecasting.

(Caveat: the suggested options below are for trial purposes only. It’s currently too early to tell which technique works best for a pandemic.)

4 ways to adapt your business forecasting techniques during a pandemic

1. The covid coefficient

If you believe the economic impact of Covid-19 will be short-term for your industry, the key is to find a way to abstract the severity of the impact into some kind of predictor in your business forecasting model. A predictor that “switches on” while the impact is present, and “off” after the impact diminishes and things go back to normal. A “covid/ no covid” indicator, if you will. 

For example, you could look at the number of coronavirus cases per day, the rate of change of cases, or a binary variable indicating lockdown. 

“In simple terms, you’ve got your basic Y as a function of X” regression analysis equation,” Kevin explains. “And you’re adding a coefficient that starts as a high number when the pandemic hit hardest, then gradually reduces until you almost go back to what it was before.”

Here’s an example of how to adapt a basic autoregressive model solving for a future value (X_{t+1}) — next month’s sales, for example. The model looks at sales from the past few months to predict what could happen next month. 

X_{t+1} = a_t * X_t + a_{t-1} * X_{t-1} + a_{t-2} * X_{t-2} + .. + a_{t-k} * X_{t-k} 

Adapting the model with an indicator representing the severity of Covid-19’s impact on your business (N_t) plus a coefficient controlling the effect of the indicator on the forecast (beta_t) would look something like this:  

X_{t+1} = a_t * X_t + a_{t-1} * X_{t-1} + a_{t-2} * X_{t-2} + .. + a_{t-k} * X_{t-k} + beta_t * N_t + beta_{t-1} * N_{t-1} + beta_{t-2} * N_{t-2} 


It all depends on how you view the lasting impact of Covid-19 on your industry. “If you think your industry is going to be impacted for the long term,” Kevin adds, “you’ll have to reevaluate your regression analysis equation completely.”

One thing to note here is that adding regressors to your model may only affect the linear trend portion of the forecast. It’s still important to consider seasonality just as you would normally. 

2. The rubber duck adjustment

Another option for business forecasting is the rubber duck curve forecast adjustment, which divvies up the forecasted horizon into four parts:

1. Regular sales (stable),

2. Disruption (Covid-19),

3. Recovery (vaccine), and

4. The new normal (stable).

If the economic impact of Covid-19 on your business is negative, the curve looks a little like a rubber duck. If positive, the curve is inverted.

giniPredict | Rubber duck adjustment chart

Note that the recovery phase could be any length of time, and it’s also possible that the new normal comes right after disruption, skipping a recovery phase altogether.

3. The narrow view

The current rate of change calls for continuous reevaluation of monthly or quarterly budgets, sometimes even on a weekly basis. The best way to deal with this level of volatility is to narrow your business forecasting horizon to the next couple of months only.

It’s also important to keep track of shifting trends in your data, and use a shorter time period to base predictions on. For example, you could adapt your regression analysis model to give more weight to the previous one month’s data, over data from the last six months or one year.

This way, the model can adapt quickly to the constantly changing data and you can achieve more accurate business forecasting outcomes.

4. The holiday effect

For those using machine learning predictive models for business forecasting, such as Facebook Prophet, one thing to watch out for is the Covid-19 effect being treated as a seasonality trend. 

Models that fit non-linear trends with seasonality effects may automatically assume the pandemic-impacted period (for example, March to July 2020) is a seasonality trend and will therefore echo the effects in the projected outcome (i.e. for March to July 2021).

To avoid this, one option is to code the pandemic-impacted data as a holiday effect that doesn’t repeat in future. That way, Prophet will model it and not include it in the forecast. 

giniPredict | Assumed seasonality effect chart

“Whatever business forecasting technique you use, it’s important to remember that there is no one right answer for everyone,” says Kevin. “It’s about trying out different forecasting techniques and statistical tools until you find what makes most sense for your business.”


Business forecasting during a pandemic — best practices

Still, there are best practices for business forecasting during periods of high uncertainty that every company can adopt immediately.

Plan for every scenario

Regardless of which business forecasting technique you use, the best strategy is to develop multiple business budgets for different economic scenarios, so as soon as something changes, a back-up plan can be implemented without delay.

“Let’s say you have five scenarios: a baseline, a bad case, a good case, a worst case, and a best case scenario,” explains Kevin. “As things progress, you’re measuring actual data against these five forecasted lines to see which is the most likely to happen.”

“When uncertainty levels are as high as they are now, it doesn’t make sense to make just one assumption. You need to look at the entire range of what could happen. As things evolve, the range begins to narrow and you start to see a clearer picture of where your company is headed, so you can plan accordingly.”

giniPredict | Scenario forecast chart

Up your business forecasting frequency

As the economic impact of Covid-19 changes, businesses need to react quickly. In this environment, budgets and predictions quickly become outdated and need to be continuously reevaluated. 

It’s important to adapt and harness technology that will automate and speed up your business forecasting processes, allowing you to quickly update plans as soon as new information arrives.

If you haven’t considered using machine learning in your business forecasting processes, now’s the time to start. Machine learning models are capable of crunching vast quantities and a large variety of data in a fraction of the time it would take a human with a spreadsheet. 

“The industries that stand to gain the most from machine learning are those that experience high levels of uncertainty,” says Kevin. “For companies that have a high proportion of variable costs — like the airline industry, for example, where the price of oil is highly volatile — machine learning comes in handy, because the models are designed to examine the rate of change of and between each variable.”

“For many years, this kind of technology was available only to industry giants like Amazon, who can afford to hire large data teams to build custom models for this kind of analysis,” Kevin continues. “But now, with the introduction of low-code solutions, more and more businesses are able to access it and benefit from it.”

Finding the right forecasting technique for your business

Above all else, the most important thing to consider when adapting your business forecasting techniques is to use what you know about your business. 

You may not be an expert in all the modelling techniques and statistical methodologies out there, but you are an expert in your business. Ensure the business forecasting techniques and tools you use line up with what’s actually happening in your company and industry right now. 

giniPredict is a business forecasting spreadsheet add-on with machine learning predictive models built in. It is specifically designed for business leaders who aren’t data scientists, so you don’t need any coding or machine learning experience to use it. With giniPredict, businesses of all sizes can benefit from the speed and accuracy of machine learning and make smarter decisions with their data. Get in touch for more.


Blu Ltd offers artificial intelligence consulting services to help you navigate the ever-changing technological landscape and bring you closer to your customers than you ever thought possible. We bring an unmatched team of experts on the frontiers of multiple streams of AI, combining business and technology acumen to take you to the next level. Get in touch for more. 


Contact us
Case study

The covid coefficient: Adapting your business forecasting techniques for a pandemic

Case study

The covid coefficient: Adapting your business forecasting techniques for a pandemic

The results
A major international bank found gini to be the most efficient data enrichment provider for its digital banking upgrade initiative.

In a pilot project with gini, the bank ran 50,000 transactions through our data enrichment engine.  Within 72 hours, gini had enriched 95.7% credit card transactions and 92.7% EPS transactions. 

“We were surprised just how fast gini’s enrichment capabilities are. What we expected to take 3 weeks took them only 3 days,” said the bank’s Head of Innovation and Strategy. “On top of that, they even enriched EPS transactions, which no other provider has achieved.”
Credit card
transactions enriched
EPS
transactions enriched
The results
gini introduced a successful Savings Goal feature in our PFM app that was adopted by 60% of users within 30 days of launching.
 
Our users engage with the Savings Goal feature an average of 7.4 times a month, which when compared to the once-a-month engagement of most banking apps, is a testament to its value. 

And the reviews were overwhelmingly positive, with comments such as, “Congrats on the release of the saving function, it’s very helpful and motivates me to save more!” and “Makes saving and budgeting a lot easier.”

Makes saving and budgeting a lot easier.
The challenge
Our research showed that users wanted a savings feature that automates their budgeting calculations, and shows how much they have left to spend after putting their savings aside every month.

However, no PFM apps in Hong Kong had a feature like this because it requires complicated algorithms and enriched transaction data. Without merchant names for example, it’s difficult to label recurring transactions accurately, and give the user a clear, comprehensive overview of their finances.

The solution
With data automatically enriched by our machine learning models, gini was able to build a fully functioning Saving Goals feature that resonated with users and increased engagement.

The new feature automatically calculates a monthly OK to Spend amount by subtracting the user’s total monthly expenses (past and upcoming) and Savings Goal from their total monthly income. It also has a traffic light system that warns users when it’s time to reign in their spending.

None of this was possible without first enriching the transaction data with accurate merchant names and categories.
The challenge
A recent digital banking survey showed low levels of satisfaction, with 87% of customers finding it hard to understand their transaction feeds.
My current spending history is confusing. I want to see the ACTUAL shop name.
To address this — and reduce queries — the bank planned to first replace standard transaction codes with clear merchant names and categories throughout its digital banking services. And then to increase loyalty with a personal finance app, built on the foundation of enriched data. 

However, developing the technology to transform such large volumes of transaction data was proving to be a Herculean task — one that would take years. So they looked for an external provider to help clean, structure and enrich the data accurately and quickly.

The solution
Impressed by the quality and speed of gini’s enrichment engine in the pilot project, the bank plans to integrate our scalable software into their own systems to allow for real-time data processing and enrichment. The best part is, gini’s technology is easily accessible as a SaaS solution on AWS Marketplace, avoiding the need for lengthy tech stack integration processes.

Soon, the bank’s entire customer base will have their transaction feeds transformed from confusing codes to recognisable merchant names, logos and categories. This is predicted to have a significantly positive impact on NPS scores.

Equipped with enriched data, the bank’s development team will then be able to build a competitive personal finance app with much richer features than otherwise possible.
Contact us to find out more about our digital banking data solutions
Contact us
Find out how data enrichment can help you build better PFM features
Contact us

Open banking in 2020: Are you ready?

Open banking is primed to become the new norm in Asia Pacific. But, as our research report shows, the majority of bankers in the region are not sufficiently prepared for what’s coming.

It’s time to get smart on what open banking is and how it’s expected to impact the market this year. 
gini's original research report on open banking in Asia Pacific for 2020
Download the research report
Download the open Banking 2020 research report by gini
We interviewed more than 300 finance and technology thought leaders across Asia on the industry’s readiness for open banking this year, with surprising results. 
Download our Open Banking 2020 research report to find out: 

The opportunities in store for all participants
The barriers to adoption
Who is expected to benefit most 
How institutions can generate revenue from open APIs
And more
giniPredict | The covid coefficient: Adapting your business forecasting techniques for a pandemic

The covid coefficient: Adapting your business forecasting techniques for a pandemic

By
on
Dec 15, 2020

In these times of unprecedented uncertainty, how do you plan ahead? It’s virtually impossible to know which way the economy will turn next, but your business still needs budgets and forecasts in order to function.

The gini team spoke to Kevin Pereira, university lecturer on Artificial Intelligence and Big Data, and Managing Director of AI consulting firm Blu Ltd., about how to adapt business forecasting techniques during the pandemic.

We’ve never seen anything like this before

Usually, when an economic shock happens, you’d estimate the impact by looking at what happened last time, or in similar industries or regions. But there’s nothing usual about Covid-19. 

With no comparable past, business forecasting techniques perfected over the years are suddenly no longer applicable. No machine learning model has been trained to account for something like this. So what do you do? 

“First, you need to figure out whether you think the economic impact of Covid-19 will be short-term or long-term for your industry,” says Kevin. “That’ll help you figure out which business forecasting technique makes the most sense for you.”

The way you adapt your forecasting technique depends heavily on your business, industry and use case. In this article, we’ll focus on regression analysis techniques for time series data, as that is the most common method used for business forecasting.

(Caveat: the suggested options below are for trial purposes only. It’s currently too early to tell which technique works best for a pandemic.)

4 ways to adapt your business forecasting techniques during a pandemic

1. The covid coefficient

If you believe the economic impact of Covid-19 will be short-term for your industry, the key is to find a way to abstract the severity of the impact into some kind of predictor in your business forecasting model. A predictor that “switches on” while the impact is present, and “off” after the impact diminishes and things go back to normal. A “covid/ no covid” indicator, if you will. 

For example, you could look at the number of coronavirus cases per day, the rate of change of cases, or a binary variable indicating lockdown. 

“In simple terms, you’ve got your basic Y as a function of X” regression analysis equation,” Kevin explains. “And you’re adding a coefficient that starts as a high number when the pandemic hit hardest, then gradually reduces until you almost go back to what it was before.”

Here’s an example of how to adapt a basic autoregressive model solving for a future value (X_{t+1}) — next month’s sales, for example. The model looks at sales from the past few months to predict what could happen next month. 

X_{t+1} = a_t * X_t + a_{t-1} * X_{t-1} + a_{t-2} * X_{t-2} + .. + a_{t-k} * X_{t-k} 

Adapting the model with an indicator representing the severity of Covid-19’s impact on your business (N_t) plus a coefficient controlling the effect of the indicator on the forecast (beta_t) would look something like this:  

X_{t+1} = a_t * X_t + a_{t-1} * X_{t-1} + a_{t-2} * X_{t-2} + .. + a_{t-k} * X_{t-k} + beta_t * N_t + beta_{t-1} * N_{t-1} + beta_{t-2} * N_{t-2} 


It all depends on how you view the lasting impact of Covid-19 on your industry. “If you think your industry is going to be impacted for the long term,” Kevin adds, “you’ll have to reevaluate your regression analysis equation completely.”

One thing to note here is that adding regressors to your model may only affect the linear trend portion of the forecast. It’s still important to consider seasonality just as you would normally. 

2. The rubber duck adjustment

Another option for business forecasting is the rubber duck curve forecast adjustment, which divvies up the forecasted horizon into four parts:

1. Regular sales (stable),

2. Disruption (Covid-19),

3. Recovery (vaccine), and

4. The new normal (stable).

If the economic impact of Covid-19 on your business is negative, the curve looks a little like a rubber duck. If positive, the curve is inverted.

giniPredict | Rubber duck adjustment chart

Note that the recovery phase could be any length of time, and it’s also possible that the new normal comes right after disruption, skipping a recovery phase altogether.

3. The narrow view

The current rate of change calls for continuous reevaluation of monthly or quarterly budgets, sometimes even on a weekly basis. The best way to deal with this level of volatility is to narrow your business forecasting horizon to the next couple of months only.

It’s also important to keep track of shifting trends in your data, and use a shorter time period to base predictions on. For example, you could adapt your regression analysis model to give more weight to the previous one month’s data, over data from the last six months or one year.

This way, the model can adapt quickly to the constantly changing data and you can achieve more accurate business forecasting outcomes.

4. The holiday effect

For those using machine learning predictive models for business forecasting, such as Facebook Prophet, one thing to watch out for is the Covid-19 effect being treated as a seasonality trend. 

Models that fit non-linear trends with seasonality effects may automatically assume the pandemic-impacted period (for example, March to July 2020) is a seasonality trend and will therefore echo the effects in the projected outcome (i.e. for March to July 2021).

To avoid this, one option is to code the pandemic-impacted data as a holiday effect that doesn’t repeat in future. That way, Prophet will model it and not include it in the forecast. 

giniPredict | Assumed seasonality effect chart

“Whatever business forecasting technique you use, it’s important to remember that there is no one right answer for everyone,” says Kevin. “It’s about trying out different forecasting techniques and statistical tools until you find what makes most sense for your business.”


Business forecasting during a pandemic — best practices

Still, there are best practices for business forecasting during periods of high uncertainty that every company can adopt immediately.

Plan for every scenario

Regardless of which business forecasting technique you use, the best strategy is to develop multiple business budgets for different economic scenarios, so as soon as something changes, a back-up plan can be implemented without delay.

“Let’s say you have five scenarios: a baseline, a bad case, a good case, a worst case, and a best case scenario,” explains Kevin. “As things progress, you’re measuring actual data against these five forecasted lines to see which is the most likely to happen.”

“When uncertainty levels are as high as they are now, it doesn’t make sense to make just one assumption. You need to look at the entire range of what could happen. As things evolve, the range begins to narrow and you start to see a clearer picture of where your company is headed, so you can plan accordingly.”

giniPredict | Scenario forecast chart

Up your business forecasting frequency

As the economic impact of Covid-19 changes, businesses need to react quickly. In this environment, budgets and predictions quickly become outdated and need to be continuously reevaluated. 

It’s important to adapt and harness technology that will automate and speed up your business forecasting processes, allowing you to quickly update plans as soon as new information arrives.

If you haven’t considered using machine learning in your business forecasting processes, now’s the time to start. Machine learning models are capable of crunching vast quantities and a large variety of data in a fraction of the time it would take a human with a spreadsheet. 

“The industries that stand to gain the most from machine learning are those that experience high levels of uncertainty,” says Kevin. “For companies that have a high proportion of variable costs — like the airline industry, for example, where the price of oil is highly volatile — machine learning comes in handy, because the models are designed to examine the rate of change of and between each variable.”

“For many years, this kind of technology was available only to industry giants like Amazon, who can afford to hire large data teams to build custom models for this kind of analysis,” Kevin continues. “But now, with the introduction of low-code solutions, more and more businesses are able to access it and benefit from it.”

Finding the right forecasting technique for your business

Above all else, the most important thing to consider when adapting your business forecasting techniques is to use what you know about your business. 

You may not be an expert in all the modelling techniques and statistical methodologies out there, but you are an expert in your business. Ensure the business forecasting techniques and tools you use line up with what’s actually happening in your company and industry right now. 

giniPredict is a business forecasting spreadsheet add-on with machine learning predictive models built in. It is specifically designed for business leaders who aren’t data scientists, so you don’t need any coding or machine learning experience to use it. With giniPredict, businesses of all sizes can benefit from the speed and accuracy of machine learning and make smarter decisions with their data. Get in touch for more.


Blu Ltd offers artificial intelligence consulting services to help you navigate the ever-changing technological landscape and bring you closer to your customers than you ever thought possible. We bring an unmatched team of experts on the frontiers of multiple streams of AI, combining business and technology acumen to take you to the next level. Get in touch for more. 


Contact us

Latest posts

Sign up for early access

Request access