Related Posts
More Posts
Looking to get a referral for BDO SALT practice.
I have had my medicals done for Australia Visa. Also, verified the medicals have been submitted to the department. Can someone please advise how much time will it take to finalise the visa after medicals submission?
It will be really helpful if someone can share their experiences?
Tata Consultancy
Additional Posts in Data & Analytics Consultants
New to Fishbowl?
unlock all discussions on Fishbowl.
I think this is a great example of a situation where domain understanding is going to matter massively compared to generalized techniques.
Not to say you couldn’t wind up applying some fancy techniques. But you’re going to need a strong hypothesis-driven approach to drivers, which may go significantly further upstream / beyond the scope of drivers one would have used to predict demand effectively in recent past. If you are an analytics expert but not an aviation expert, you should absolutely be getting one on the phone and brainstorming together to come up with your long list, and then considering if that requires additional types of data beyond what has been specified for you already.
Otherwise if you don’t have such driver based hypotheses, all the data you could want on you dependent variable itself and some classic correlates will probably just get you a model that still comes down to one or two crude implicit curve-fitting assumptions - even if they’re obscured by fancy math you put on top.
Bowl Leader
Your MD needs to bring in help. This is not an analysis that should be given to someone without statistical training. If you have to deliver something, stick to segmentation and simple estimation like you might do on the spot for a case.
What are you trying to predict
Be careful. Unprecedented times means there might exist data with no appropriate outcomes. Your algorithms might not generalize to the current data.
My advice is to make sure it’s not a pipe dream.
Fair warning but I disagree. Ppl are already doing it and fairly accurately.
Bowl Leader
Train only on similar periods. Finding similar periods will be contextual to your goal. Test on data from current event, make sure you cross validate.
I’m in a similar boat. It’s realllllly hard.
I’ve found that GDELT data is helping a lot. The problem is predicting media activity around COVID is tough, as negative news around the disease increases during the beginning of an outbreak, and the sentiment gets more positive as the disease plateaus and decreases as articles increase as well.
Soooo many assumptions, tweaking, etc needs to be made to these models. And they change everyday.
Good luck. I would say your best bet is using the predicted deaths from IHME (since they have a CSV dataset) and stemming from there
Is difficult to source but I feel I can get that done. The part I'm losing sleep over is how to actually model/Create a predictive model?
I've been tasked to look at past catastrophic events and use that as a basis. Not sure what that means. If anyone is an expert please help a young senior associate out. My MD is on 10 different things and just flies high with the selling. Thanks team
That sounds awesome. Any chance you’re At BAH?!
Also, What are you trying to predict? Whether it’s a binary outcome or not will def influence any further advice.
Customer demand declines impacting Airline industry revenues, when we can expect it to come back up, MRO/shop visit declines, employment declines, manufacturing (Boeing, air bus) declines.
I don't work at BAH. Not the point. I'm trying to figure out where to start after the historical data is source. And how to correlate stuff as to make a predictive model..? Just confused
Maybe I'm just over simplifying. But can't I just use this as my hypothesis?
https://www.bloomberg.com/news/articles/2020-04-08/airlines-to-cut-summer-flights-up-to-90-with-rebound-far-off
Oversimplifying is typically the best place to start. The assumptions will be obvious and people will understand the analysis limitations. As you build complexity into the analysis, you can use the oversimplified analysis as a benchmark or point of reference. This helps you build confidence with stakeholders as you help them gain insights into their business problem. At some point, when you’ve built a complex demand model trained on SARS in Asia, Ebola in Africa, and recent civil wars in the Middle East, you can 1) point back to simple assumptions, 2) how your model has refined estimates of those assumptions, and 3) why your model may be right and what would likely make your model wrong
This might give you a framework to start
https://www.nber.org/papers/w26987?utm_campaign=ntwh&utm_medium=email&utm_source=ntwg18