Use metrics to find what's a performance problem, don't just guess what it is.
As you're accessing it via a web interface, the datastore at the back end is irrelevant. The performance considerations for you concern the transport of the data to your client and then the parsing of the data into workable data structures (i.e. C# classes).
If you have control of the format you receive the data in, and it's tabular, then CSV will smaller than JSON. If not, then you have to work with what you're given.
You said you can parse from CSV into JSON, which means you already have a working CSV parser and so have no need for the JSON.
The performance hits now are to do with processing power (to read the CSV) and memory (is it big enough to hold the dataset you want to work with?
If those are fine then you need to consider the number crunching and searching. To find the data (in memory) that you're interested in you need to use a nicely indexed data structure. You might find that the
System.Data.DataTable is best, it's old and can cause allot of GC collections but it may be good enough for your MVP. If that doesn't work then you need to write a bespoke structure.
You may also was to calculate some aggregations and store those so that they're not calculated on the fly (you might even have these delivered to you by the call to your PHP file, the server & DB are far more capable of doing this that a mobile device), as well as store aggregations the user has requested ready for when they ask for them again.
Long answer, but you question isn't a simple one to answer. Whatever you decide to do, you must gather metrics in order to find out what is really the performance problem rather than what you think is the performance problem.