Tonny Menglun Kuo
Assistant Research Fellow @ National Tsing Hua University (Taiwan) https://sites.google.com/view/tonnykuo
Location Hsinchu City, Taiwan
Activity
-
Thank you Guus. I have updated it as well ;)
-
Dear Jackie,
Thank you very much for your feedback. We revised the wording of the Question to make it more clear. ;)
-
Dear Ben,
Thank you so much for your comment. We have updated it in the current article. ;) -
Dear Jackie,
This post might be helpful to your Qs: http://stats.stackexchange.com/questions/130448/how-to-undifference-a-time-series-variable
You can take a look and check it out!
-
I think the best way to identify the type of trend is to see the plot of your data. This gives you overall understanding of what the data look like. No only can you find the trend and seasonality, but you can find out some outliers or extreme values.
Of course, you can try both models to compare their differences using histogram, residual, and other...
-
You can use the visualization tool (such as Tableau) to help you understand what type of trend your data is.
-
You can use it manually. Lag1 or lag 12 is just to keep the first one value (lag 1) and the first twelve values blank.
-
Tonny Menglun Kuo replied to Ben Petschel
Yon can see differencing as a way to remove trend and seasonality. It could be used not merely in smoothing but also other methods. However, it would be great to understand how differencing works and what differencing is for.
-
Tonny Menglun Kuo replied to Isabel Du Solier
Good point! It's good to understand the business/forecasting goals and forecasting when choosing specific method for forecasting. Although MA is very simple, it offers very basic forecasting information. Sometimes, MA can bring your stakeholders satisfactory results.
-
Dear Kleyn,
You can consult this webpage for more information. :) http://robjhyndman.com/hyndsight/estimation2/ -
Tonny Menglun Kuo replied to Isabel Du Solier
In simple naive forecast, only one previous value is considered. However, in the seasonal naive, we use different mean of each part. Using all the values as mean could be seen as a benchmark, but it's not naive forecast that we talks about here.
-
Usually, we use naive forecast as baseline. However, seasonal naive could be used too.
-
Tonny Menglun Kuo replied to Isabel Du Solier
Please dont worry. There are other models that you will learn such as smoothing, regression, and ARIMA. Naive forecast is the most simple one.
-
Yes. Before running the data, you'd better ensure your data set is readable by any software you will use. Data cleaning is quite important in data analysis process.
-
I believe you can! But in time-series forecasting, since we need to partition the data into different sets and to verify the forecasting performance, I more suggest to partition using the time stamp.
-
Tonny Menglun Kuo replied to Aruna Bhatti
Good point. Considering the trend abd seanality could be very helpful, in particular in choosing different model.
-
Tonny Menglun Kuo replied to Juan Alvarez
Actually transaction data with time stamps could be seen as time series data. Therefore, time series analysis is pretty practical.
-
Great point! In many cases, we focus on both goals, rather than "predictive" or "descriptive" alone. However, knowing your goal is quite important for your forecasting/analysis!
-
Try to stay focused on "helping your stakeholders understand the value of forecasting". Using the clues in the video to ask related Qs will brings better understanding to the stakeholders.
-
Think about our step title "Working with a stakeholder to discover forecasting opportunities". In many cases, your clients/stakeholders might have little knowledge of "forecasting". How to transform your forecasting Qs into "their language" and help them understand the value of forecasting would be a good point here!
-
Your questions are clear. According to the video, try to use "if..., ...." clause to help the manager to figure out what they and you (as a business analyst) can do.
-
Taking to your stakeholders, especially for those who have limited experience in forecast, try to avoid "technical terms". Using plain words with examples that meet their business goals boost their willingness of communication.
-
Cool! But try to think about "guiding the manager to understand the power of forecasting". Not just to develop new stuffs. Remember, once the menu is changed, you will not have any previous data to forecast.
