What is the trickiest thing to get right in Rates Quant recently (2019)?

Question

What are the biggest challenges for Rates Quants in 2019? Most quants have been through a lot over the past years-shifting their SABR models in JPY swaptions, fixing the FVA models for negative rates, adjusting to the new cash formula for EUR swaptions, dealing with a multitude of discount rates depending on the CSA, adding copulas to midcurves. Most everyone knows how to change vols to deal with high strike payers/CMS. SOFR is looming on the horizon and IBOR-OIS may be a nonissue for many countries soon enough.

So, what is the biggest problem facing Rates Quants now? What will 2019 bring as the biggest sets of challenges?

(Preferably not the pricing of some asianed collared CMS spread trade...something more fundamental).

Answer 1

Of course making money is always the key issue. That (not completely facetious) comment aside:

1. On the practical side, in many firms IT is struggling with being clear, transparent, and intuitive in their handling of multiple curves and their associated risks. Stumbling over your own systems is an annoying way to lose money. These risks can be surprisingly large under market duress, when banks have to raise money in their native ccy to support paper in, say, USD.
2. I think the CMS/high strike issue will go away shortly; we have an alternative method for handling convexity corrections which doesn't involve replication and the consequent high strike risks ... should be available in couple weeks. Should also handle cash settled swaptions with the settlement based on IRR.
3. There's the perpetual problem of managing forward volatility surfaces / smiles / risks. Deterministic vol models are poor, stochastic vol models (Heston, SABR) are better at foward volatilities, but are poor at predicting forward volatility smile. So are stochastic/local volatility models ... in fact, post-calibration they usually give very flat volatility smiles. Very annoying.
4. in the current low rate environment, many of our desks will be turning to credit and structured products for better opportunities. How can we use our unique skills in the fixed income market (e.g., our understanding of smiles) to find opportunities among the tranches? Or maybe X-Ccy products?
5. And then the elephant in the room. Currently we can ignore machine learning. It's 50% crap, 47% same-old-thing with fancy new titles, and 3% really, really interesting stuff. In ten years most of our work will be heavily predicated on MI. (It's everywhere, even in the Mafia: people who robbed this bank also enjoyed robbing ...). Since I have no desire to retire in 10 years, how do we fund our transition, where we learn how to make MI work effectively for our markets? Since we cannot make money doing what everybody else does, I think we need to use the statistical clustering algorithms on derived quantities, where low dimensionality may be more apparent. I think there is too much emphasis on high frequency trading; this is like trying to harvest the ripples on top the water, where the real money is probably in understanding the tides. I also think there is too much emphasis static methods, adding noise to the data and using adaptive methods may be more effective at seperating the wheat from the chaff ...

Anyway, apologies for this stream-of-consciousness posting