RStudio<\/strong> (not R \u2014 look for the RStudio icon, which has a blue circle with an „R“ in it). You should see four panels. The bottom-left one is the Console \u2014 that’s where you’ll type your first commands.<\/p>\nIf you see four panels and a blinking cursor, congratulations. You’ve done the hard part. Everything from here is just typing.<\/p>\n
Step 3: Install fpp3<\/h3>\n
In the RStudio console (bottom-left panel), type this and press Enter:<\/p>\n
install.packages(\"fpp3\")\n<\/code><\/pre>\nR will download about 70 packages. This takes 2\u20135 minutes depending on your internet speed. You’ll see a lot of text scrolling by \u2014 that’s normal. It’s downloading the tidyverts, core tidyverse components, and all their dependencies. Go refill your coffee.<\/p>\n
Two things that might happen (and shouldn’t worry you):<\/strong><\/p>\n\n- If R asks „Do you want to install from sources the package which needs compilation? (Yes\/no\/cancel)“<\/em> \u2014 type
no<\/code> and press Enter. The pre-built version works perfectly fine.<\/li>\n- If R asks you to choose a CRAN mirror \u2014 pick any one close to your location. They all have the same packages.<\/li>\n<\/ul>\n
When it’s done, you’ll see the cursor blinking again with no error messages. If you see red text that says Warning<\/code> \u2014 that’s usually fine (warnings are suggestions, not failures). If you see red text that says Error<\/code> \u2014 something went wrong. The most common fix: close RStudio, reopen it, and try again. The second most common fix: update R to the latest version from CRAN.<\/p>\nTo verify everything worked:<\/p>\n
library(fpp3)\n<\/code><\/pre>\nIf you see a message listing the loaded packages (tsibble, fable, feasts, and friends) with no errors \u2014 you’re in. The entire professional-grade forecasting toolkit is now on your machine. For free. While your ERP vendor is still emailing you about their „advanced analytics module“ that costs six figures.<\/p>\n
Your First Session: From Zero to Forecast<\/h2>\n
Let’s use that toolkit. We’re going to load a real dataset, visualize it, and produce a forecast with prediction intervals \u2014 all in about ten lines of code.<\/p>\n
Look at Real Data<\/h3>\n
The fpp3 package comes with dozens of built-in datasets from real sources. We’ll use aus_production<\/code>, which contains quarterly Australian beer production in megalitres. Why beer? Because it has beautiful seasonality (Australians brew more beer in Q4, their summer), a clear long-term trend, and it’s the kind of production planning data supply chain professionals deal with every day.<\/p>\nAlso: beer.<\/p>\n
library(fpp3)\n\naus_production |> select(Quarter, Beer)\n<\/code><\/pre>\nThe |><\/code> symbol is called a pipe<\/strong>. Read it as „and then.“ So this line says: „Take aus_production<\/code>, and then<\/em> select the Quarter and Beer columns.“ If you’ve ever chained Excel formulas inside each other like =ROUND(AVERAGE(IF(...)))<\/code> and wanted to scream, pipes are the readable version of that idea.<\/p>\nMake Your First Plot<\/h3>\n
Here’s where it gets fun:<\/p>\n
aus_production |>\n autoplot(Beer) +\n labs(title = \"Australian Quarterly Beer Production\",\n y = \"Megalitres\")\n<\/code><\/pre>\n![\"Your](\"https:\/\/inphronesys.com\/wp-content\/uploads\/2026\/04\/intro_first_plot.png\")
<\/p>\n
One line of actual plotting code \u2014 autoplot(Beer)<\/code> \u2014 and R gives you a publication-quality time series chart. The seasonal pattern jumps off the screen: production peaks every Q4 (Australian summer) and dips every Q2 (winter). You can see the boom years through the mid-1970s, the structural shift as drinking habits changed, and the gradual decline into the 2000s. Decades of production history, instantly visible.<\/p>\nIn Excel, getting a chart this clean would take you ten minutes of formatting: adjusting axes, removing gridlines, fixing the date labels, fighting with the legend placement. In R, it took one function call.<\/p>\n
Run Your First Forecast<\/h3>\n
Now the moment you’ve been waiting for. Let’s take the data from 1992 onward (the more stable recent period) and ask R to forecast three years ahead:<\/p>\n
beer <- aus_production |>\n filter(year(Quarter) >= 1992) |>\n select(Quarter, Beer)\n\nfit <- beer |>\n model(ETS(Beer))\n\nfit |>\n forecast(h = \"3 years\") |>\n autoplot(beer) +\n labs(title = \"Australian Beer Production: 3-Year Forecast\",\n subtitle = \"ETS model with 80% and 95% prediction intervals\",\n y = \"Megalitres\")\n<\/code><\/pre>\n![\"Your](\"https:\/\/inphronesys.com\/wp-content\/uploads\/2026\/04\/intro_first_forecast.png\")
<\/p>\n
That’s it.<\/strong> Five lines of real code, and you have a forecast with:<\/p>\n\n- A seasonal pattern<\/strong> that R detected automatically \u2014 peak in Q4, trough in Q2, year after year<\/li>\n
- Prediction intervals<\/strong> \u2014 the shaded bands showing where production is 80% and 95% likely to fall<\/li>\n
- A named model<\/strong> \u2014 this is an ETS(A,N,A) model: Additive errors, No trend, Additive seasonality<\/li>\n<\/ul>\n
Let’s unpack what each line did:<\/p>\n
\n\n\n| Line<\/th>\n | What It Does<\/th>\n | Excel Equivalent<\/th>\n<\/tr>\n<\/thead>\n |
\n\nfilter(year(Quarter) >= 1992)<\/code><\/td>\n| Keep only data from 1992 onward<\/td>\n | Deleting old rows manually<\/td>\n<\/tr>\n | \nselect(Quarter, Beer)<\/code><\/td>\n| Keep only the columns we need<\/td>\n | Hiding columns<\/td>\n<\/tr>\n | \nmodel(ETS(Beer))<\/code><\/td>\n| Fit an ETS forecasting model<\/td>\n | There is no Excel equivalent<\/td>\n<\/tr>\n | \nforecast(h = \"3 years\")<\/code><\/td>\n| Predict 3 years into the future<\/td>\n | =FORECAST.LINEAR (but worse)<\/td>\n<\/tr>\n | \nautoplot(beer)<\/code><\/td>\n| Plot forecast with intervals<\/td>\n | 30 minutes of chart formatting<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n The line that matters most is model(ETS(Beer))<\/code>. That single function call estimates the optimal smoothing parameters, selects the best error-trend-seasonality combination using the Akaike Information Criterion, and fits the model to your data. Behind the scenes, R just evaluated 30 candidate models and picked the winner. Excel doesn’t even know those 30 models exist.<\/p>\nWhat Just Happened?<\/h2>\nTake a breath. You just:<\/p>\n \n- Installed a professional statistical computing environment<\/li>\n
- Loaded a curated forecasting toolkit used by thousands of practitioners worldwide<\/li>\n
- Visualized a real production time series with one line of code<\/li>\n
- Produced an honest-to-goodness ETS forecast with prediction intervals<\/li>\n<\/ol>\n
And you didn’t write a single VLOOKUP. Not one INDEX(MATCH()). No circular references. No broken chart axes. No mysterious #REF!<\/code> errors cascading through seventeen worksheets.<\/p>\nIf you’re feeling a mix of „that was surprisingly easy“ and „I have no idea what ETS actually does“ \u2014 that’s exactly where you should be. You don’t need to understand the mathematics of exponential smoothing to use it effectively, just like you don’t need to understand combustion engineering to drive a car. The understanding will come with practice. For now, the important thing is that it works<\/em> and you made it work<\/em>.<\/p>\nYour Turn<\/h2>\nDon’t close RStudio. Try these:<\/p>\n 1. Change the forecast horizon.<\/strong> Replace \"3 years\"<\/code> with \"5 years\"<\/code> or \"1 year\"<\/code>. Watch how the prediction intervals widen as you forecast further out \u2014 that’s R being honest about increasing uncertainty. Excel’s linear forecast doesn’t get wider. It should.<\/p>\n2. Try a different dataset.<\/strong> The aus_production<\/code> table also has Cement<\/code>, Electricity<\/code>, and Gas<\/code> columns. Replace Beer<\/code> with any of them:<\/p>\naus_production |>\n autoplot(Cement) +\n labs(title = \"Australian Cement Production\", y = \"Tonnes ('000)\")\n<\/code><\/pre>\n3. Compare two models.<\/strong> This is where R starts to pull away from everything else. (If you’ve restarted RStudio since the forecast section, re-run the code from „Run Your First Forecast“ first to recreate the beer<\/code> variable.)<\/p>\nbeer |>\n model(\n ETS = ETS(Beer),\n ARIMA = ARIMA(Beer)\n ) |>\n forecast(h = \"3 years\") |>\n autoplot(beer, level = 80) +\n labs(title = \"ETS vs ARIMA: Who Forecasts Beer Better?\",\n y = \"Megalitres\")\n<\/code><\/pre>\nTwo models, one chart, six lines. Try doing that in Excel. I’ll wait.<\/p>\n 4. Check the numbers.<\/strong> Want to know which model actually wins?<\/p>\nbeer |>\n model(\n ETS = ETS(Beer),\n ARIMA = ARIMA(Beer),\n SNAIVE = SNAIVE(Beer)\n ) |>\n accuracy() |>\n select(.model, RMSE, MAE, MAPE)\n<\/code><\/pre>\nThis prints a clean table comparing Root Mean Squared Error, Mean Absolute Error, and MAPE across all three models. No manual calculation. No helper columns. No pivot tables. Just the answer.<\/p>\n What’s Next<\/h2>\nYou’ve got the tools installed and your first forecast on screen.<\/p>\n Next, we go deep. We’ll take a real multi-SKU supply chain dataset through the complete fpp3 workflow: cleaning messy data, decomposing time series to understand their hidden structure, fitting multiple models, cross-validating properly (not just eyeballing the chart and hoping for the best), and comparing accuracy metrics the way statisticians actually do it. If today was „start the car and drive around the block,“ next time is „take it on the highway.“<\/p>\n You’ll walk away with a reusable R script template you can plug your own ERP data into \u2014 the same workflow I use for production forecasting.<\/p>\n If you haven’t installed R yet: do it now. Not tomorrow. Not „when things calm down“ (things never calm down \u2014 you work in supply chain). Right now. The install takes 10 minutes, and everything we cover from here on out assumes you have a working setup.<\/p>\n And if you get stuck \u2014 on installation, on a confusing error message, on anything \u2014 drop it in the comments. I’ve debugged more R install issues than I’ve debugged supply chain disruptions, and I’ve debugged a lot of supply chain disruptions.<\/p>\n See you next time.<\/p>\n
| | | | | |
![\"The](\"https:\/\/inphronesys.com\/wp-content\/uploads\/2026\/04\/intro_fpp3_ecosystem.png\")
<\/p>\n