Improve QRA

[nikosbosse]

A few current issues with the QRA:

• failure to run with too few data points or too many models. This leads to infite runtime
• sometimes no sensible weights are estimated

What we could do:

• look through what is happening in detail and why
• set a maximum to the quantgen runtime to make sure it doesn't run indefinitely
• look at the grid of horizons that go into QRA and make sure there are always enough data points to estimate weights
• implement a flag that alerts us if something goes wrong when computing QRA weights
• implement an automatic fallback to equal weights (or the NULL model?) whenever something goes wrong

[seabbs]

Related to #91

[seabbs]

Another option is to change package or method. For example quantreg.

[sbfnk]

And/or raising these as issues in https://github.com/ryantibs/quantgen