Forecasting Methods

You may have heard about it a thousand times:

• “We have a predictive model that tell us… bla, bla, bla…”.

Having a predictive forecasting model is one of the most important tools you should have.

However, wherever you look for a “predictive” model for your sales, revenues, pageviews… you find very difficult explanations and advanced mathematics.

Here, we want to share with you how to develop your own simple predictive models in a simple, useful and understandable way.

But first of all, we have to explain you what a forecast model is, why it is so important, and its limitations:

A Forecast model is a mathematical model that tries to predict how certain outcome will evolve in the future according to its past trend.

The more variable involved in a process, the more difficult the forecast model would be.

As you may have already guessed, there are simple forecast models extremely complex ones.

It is probably the most simple and easy to understand forecast model:

If you flip a coin in the air, which side of it will come out?

Assuming you flip the coin in a perfectly “random” way, there are just 2 options with exactly the same probability (50%):

• Heads and Tails.

It may seem very obvious, but, with this simple predictive model you can guess what are the chances of having 3 consecutive “heads” as result, for example:

• 50% x 50% x 50% = 12.5%.

This is not that “intuitive”, isn’t it?

If you were betting with somebody that had no idea about maths or statistics, you could earn some money by comparing probabilities and analyzing the chances.

* Of course, you could develop an extremely difficult forecast model, taking into account:

• How you flip the coin.
• Wind (for saying something).
• The type of coin you use.
• etc.

Maybe, you would discover that one side is a little bit more likely to come out.

This example shows how sometimes, from a simple and basic forecast model, we can get more complex not-so-intuitive information.

This is “the other side of the coin” (it would have been impossible to pick a better expression).

Meteorological forecast models are extremely difficult to design.

Think about all the variables involved:

• Wind speed.
• Temperature.
• Air density.
• Air viscosity.
• Country/ City/ Town height.
• Mountains or any other geographical element that could affect air currents.
• Earth rotation and Coriolis effect.
• etc, etc.

Some of the most powerful computers in the world take thousands of variables into account in order to predict the weather and they can only predict 3 days forward “accurately” (sometimes not even that).

As you may have seen with these examples, we all use forecast models everywhere in our day a day.

But why is it important to use forecast models in our professional projects?

Can we really trust these forecast models?

You may think that trusting a “model” that may commit mistakes makes no sense.

Most of the people and Business owners think this way (we now from our experience).

Using the previous “coin” model, someone could say:

• “I’ve flipped the coin 3 times and I got 3 “heads” in a row, so 12.5% probability seems not very accurate”.

However, if you kept flipping the coin 3 times in a row, let’s say, 1.000 times, you would appreciate how accurate that 12.5% predicted is.

Forecast models predict how an outcome will evolve with a certain error margin.

But, with a good forecast model you will make fewer mistakes than you would make without it.

• Moreover, a good forecast model is open to modifications and improvements, as we’ll explain later.

As you know, we have experience in Venture Capital.

We’ve analyzed and restructured lots of companies that have been under “terrible” management.

And one of the most common mistakes is how to estimate next year’s sales.

• Based on that sales, you have to make a budget, schedule investments… So, it is very important.

Main mistakes when “forecasting” next year’s sales:

Adding a fixed percentage increase:

• You would never imagine how often, lots of companies estimate their next year’s sales just adding a fixed 5% increase. That is all.

Being unjustifiably optimistic:

• On the other hand, when a certain company’s sales are plummeting, new hired commercials tend to make extremely optimistic sales forecasts.
• Sales, then, adopt a “V” evolution.

If you have the same product, with the same quality, and you’ll sell mainly in the same market, your sales won’t skyrocket miraculously.

Assuming the same sales than the previous year:

• This always ends up in lower sales.
• If a company is going through a bad situation, lots of managers will assume that sales will remain constant.

Sales never remain constant.

We don’t say that, with a certain forecast model you will stop committing mistakes. Of course you will commit mistakes, but at least you’ll be aware of its limitations and hopefully, that mistakes will be smaller than the ones you would commit without that forecast model.

If forecasting models can be that helpful, why people don’t use them more often?

There are 3 main reason people don’t use forecasting models:

• They are never completely accurate.
  • As we commented before, there is no perfectly accurate forecast method.
  • There is a curious and common paradox: Managers prefer their non-accurate estimations to more developed models because they are used to them.
    • They reject forecast methods due to their “inaccuracy” when they trust their non-supported forecasts.

• They must be periodically re-viewed.
  • They require periodic checks in order to find out whether they are accurate enough or they could be improved somehow.

• People are afraid of using maths.

* This last reason is more common that you would think.

That is why we’ll propose you:

• Easy understandable mathematical methods and models.
• Guidelines for using them.

We’ll now propose you different models depending on:

• What you are trying to forecast.
• The market or sector you’re in.
• Whether or not, you have previous data.

Let’s begin:

This model is one of the most simple among all. Nevertheless, few people use it in a proper way.

• When you have no previous data.
• When past results have increased (or decreased) in a constant way.
• When past data has evolved without any clear tendency.
• When you are estimating sales (in order to be conservative).
  • Unless you have an astonishing product that has proven to be sold with exponential increases.
• When the sector or market you’re forecasting is growing moderately at standard rates (lower than 3%, for example).

This is the most common and sensible way for forecasting outcomes.

They key is to use good representative data and analyze your initial conditions.

We’ll then propose 2 main scenarios for using this method:

1. Starting from scratch scenario.
2. Forecasting with previous data.

In this scenario, we’ll show you how to forecast your results when you start from scratch.

When you forecast without previous data, you need to rely on 2 main pillars:

• Average market growth.
• The results that can be realistically achieved.
  • If you were selling handmade soap, for example, you shouldn’t expect to sell 10.000 soap bars per month if you only could manufacture 300.

This mathematic equation is the one that will guide you through your forecasts.

What does it mean? 

y = the outcome sales, demand, revenues, pageviews, users, etc.

x = time (most usual); months are a relatively good time period.

m= the slope; the more inclined, the more rapid the outcome will increase.

n = starting point; your “zero monthly sales”. What you can predict almost for sure.

Analyzing the data carefully, developing a proper Pareto Analysis is vital for identifying your 20% “space”.

Let’s see this with a helpful example:

Imagine that you are planning to start a handmade Jewelry Business.

You have already sold different products and now you want to know how much you would be able to earn.

* We could do a more precise analysis with everything you would need, how should you increase the business, your margins and costs… but we’ll focus on the amount of sales you’d be able to supply (that simplifies everything).

You know that you could easily make $1.500 per month, but you don’t know how much time would it take you to reach $4.000 per month.

Let’s assume:

• You have the “technical capacity” for reaching $4.000 per month and even more.
  • You need time for improve your branding.
• The world average growth is: 3% (according to IMF).

You are missing one thing: handmade jewelry sector average increase rate.

You decide to check Google Trends and consider the “Interest” evolution as the average growth for this market (a relatively fair assumption).

• Visit our “how to use Google Trends” Page and find out how to properly use Google Trends.

You find out this:

The interest/growth is stable or even decreasing.

However, this doesn’t mean that, with the proper product you wouldn’t be able to penetrate the market.

You then decide to take a conservative approach and use the “Global” average growth rate; 3% as your expected growth rate.

How would your model look like?

• n = 1.500
• m = Growth rate estimated * Initial starting point = 3% * 1.500 = 45.
• y = 45 * x + 1.500 = Estimated sales after x months.
• x = (y-1.500)/45 = Estimated months you need for y sales.

And when would you achieve that $4.000 sales?

• According to our Lineal model, after 56 months. 4 years and a half.

Warning

This model assume constant increases: (Growth rate considered) * (Starting point) = 45 in this example.

A constant percentage increase model is not lineal, since each month, the difference increases in absolute terms.

• It would be exponential/ logarithmic: We’ll explain it later.

This “linearization” is very useful since:

• It simplifies the numbers and the calculations.
• It is more conservative.

However, keep in mind that the further the forecast, the lower the precision would be.

In this scenario, we’ll assume that you have already started your Business or Project.

Based on your past results, you’ll have it much more easy to justify your expected growth since you’ll be forecasting your real tendency (for better or worse).

You’ll look for the same equation but you’ll let Excel to build it for you:

Now, we have assumed the same scenario with a little difference:

• This time, you have been a whole year selling your products and now, you want to find out, after this whole year, when would you be able to reach $4.000 monthly sales.

Again, we assume you have enough technical capacity for covering all your demand.

As you can appreciate, the equation resulted (calculated by Excel) is very similar to the previous one.

* Of course, we used similar numbers.

When would you reach $4.000 sales per month?

• Reversing the equation: x = (y – 1.452)/44,5
• You would need 57 months. 4.7 years.

Why are these 2 examples important? Isn’t all of this obvious?

No. It is not that obvious.

We have explained the most basic and “conservative” models.

Although it makes sense to check past sales and expect a linear forecast (such as we have done in our second example) sales managers rarely do it.

• You would never imagine, how often professionals expect their sales to increase exponentially. Year over year.

Now, we’ll explain the “exponential” model and you’ll be able to compare the time expected to reach those $4.000 per month from the examples:

This is, generally speaking, a much more “optimistic” scenario.

Maybe you are wondering yourself: “How is it possible? Nobody make exponential forecasts”.

If you think so, you are wrong:

Assuming a constant percentage increase means exponential forecasts.

Let’s analyze these forecasts as we did before:

• Assuming you have no previous data.
• Assuming you have data from past activity.

Again, we’ll need 3 main variables for this model:

• The Starting Point.
• The Growth.
  • Ideally is the sector growth.
  • If you don’t have the growth of the sector, you can use the growth estimated for the economy you’ll be selling at.
    • Worlwide, Europe, USA, etc.
• Technical capacity.
  • Don’t expect what it is impossible.

Don’t be afraid.

You probably have been doing this mathematical operation for long time, even if you don’t know it.

What does it mean? 

y = the outcome; sales, demand, revenues, pageviews, users, etc.

xi = initial Starting point; zero month sales (it was “n” in the last expression).

p = growth rate considered in base-1 (5% for example is 0.05).

n = periods of time considered where you thing “p” takes place; Months , years…

• For example, if you consider that your Business would grow at a 5% rate per year: n = years.

However, there is a usual problem with this expression:

• While the “linear” approach was pretty easy to “revert” (find “x” instead of “y”) this one is a bit more difficult.

How could you find the time necessary for reaching a certain outcome?

How could you guess “n”?

The expression that would allow you to calculate the time necessary for reaching a certain outcome is:

With a simple calculator, it is a very easy equation.

So, if you are estimating percent increases (exponential) and you want to know when you would reach a certain number, this is now your expression.

Let’s analyze the same example than before, with this new mathematical model:

We’ll assume the same scenario as before:

• Initial sales: $1.500.
• Global growth: 3%.
  • We’ll consider this growth again for our monthly sales’ increases.

Resulting exponential model:

Sales would follow the next tendency:

Although this tendency seems linear, it is not.

While the linear approach increased the sales in $45 each month (3% of the first month sales) this approach increases a 3% of the previous month.

• In absolute terms it increases each month.

Let’s find out how much time you would need for reaching $4.000 per month:

You would need 33.2 months for reaching $4.000 monthly sales.

• Remember, in the Linear model you needed 56 months.

However, what if you had previous data?

How could you build an “exponential” model?

In this case, since the mathematics are quiet complicated, we’ll rely again on Excel (we don’t know what would we do without it).

Again, we’ll assume that you have data from past activity.

Moreover, you should have appreciated a remarkable increase in sales for considering this model and not the Linear one.

We’ll tell you how to handle this model with a practical example, using the same jewelry scenario:

For this example we have assumed the same sales that we had in the “Linear with past data” scenario.

• This way, we’ll appreciate the differences between these 2 models clearly.

* Remember: We want to find out when would we reach $4.000 sales per month.

We created an exponential trend line with Excel (base 2).

But again, we have to “revert” the equation so we can easily find x (here, x is our n):

If “y” = $4.000, this model tell us that our sales will reach this sales in 17 months.

As you may have already appreciated, all these models predict a quite different outcome.

Then, which model should you use?

Let’s see again the results we obtained in our examples:

Analysis without previous data:

• Linear model: 56 months for reaching $4.000 per month.
• Exponential model: 33 months.

Analysis with previous data:

• Linear Excel model: 57 months.
• Exponential Excel model (base 2): 17 months.

Now you probably understand why some people don’t use these mathematical models:

• They are afraid of maths.
• They don’t really understand the models.
• Besides, these models differ in the results predicted with just modifying a single variable.

Depending on which one you choose, the outcome predicted vary significantly.

What we suggest you?

Develop 2 scenarios:

• An optimistic scenario, with an Exponential model.
• A moderate scenario, with a Linear model.

Then, as soon as you start having new results, you’ll be able to check whether you can expect one scenario or the other.

In our examples, you could conclude:

• We won’t be able to achieve $4.000 per month even if our sales increased exponentially.

* In reality, you could reach those $4.000 per month in 1 month, o course; if you released the best product in the market at the best price…

• Forecast models work for average scenarios.

Don’t stick just to one forecast model.

This way, you can set milestones and goals to achieve.

A Forecast model is a very useful tool that allows you to analyze future outcomes or results.

There are an unlimited amount of forecast models, depending on what you are analyzing.

We propose you 4 simple but powerful forecast models:

• Linear without past data forecast model
• Linear with past data forecast model.
• Exponential without past data forecast model.
• Exponential with past data forecast model.

With these forecast models you’ll be able to:

• Analyze different scenarios.
• Set milestones.
• Find out what you could expect.
• Assess how volatile your market is.

Don’t be afraid of the math involved and anticipate future events waiting for all possible scenarios.