A steepening curve is no haven for agency MBS

Agency MBS finished July down 1.42% on the Bloomberg US MBS Index, and the widening that produced the loss is the part worth sitting with: 7 basis points to 31 OAS, an excess return of -0.44% against Treasuries. Investment-grade corporates lagged Treasuries by 0.26% and high yield by 0.08%, according to Pioneer Investments portfolio manager Tyler Patla, making agency MBS the clear laggard, and the underperformance was fairly even across lower and higher coupons.

Patla, a managing director and deputy director of core fixed income who runs agency mortgages at Pioneer, describes July as a month of market-moving headlines with no single theme. The tenuous June ceasefire between the US and Iran eroded, oil spiked, confidence in AI capital spending wavered, and long-term Treasury yields extended higher after Fed Chair Warsh's second post-FOMC press conference left the curve steeper and inflation expectations higher. The month closed with the US Treasury intervening alongside Japanese authorities to slow the yen's fall, even as the economic data stayed reassuring: solid growth, cooling monthly inflation, and robust second-quarter earnings.

Agency MBS oscillated through it all and sold off in the final days, extending what Patla calls the sector's tendency to trade to longer durations than model predictions in a selloff. The widening happened while the curve was steepening, a combination mortgage OAS is built to preclude: it controls for expected volatility and rates, including the yield curve's shape. Patla's note splits every month of the past 15 years into four buckets based on whether 2s10s steepened or flattened and whether rates rose or fell, and the table, built from Bloomberg and Pioneer data as of July 31, shows OAS has tended to widen in some rate environments and tighten in others. July, he says, was an unlikely response given that precedent.

After Warsh's hawkish June meeting, this page argued that the carry in agency MBS is thin; June's 2-basis-point widening on a flatter curve set the table, and July's 7-basis-point move on a steeper one confirms it. The 31-basis-point OAS looks cheap in isolation, but Patla's framework says the level matters less than the environment: when the same spread move arrives with a steepening curve and a Fed chair actively reshaping inflation expectations, the sector is no longer a static spread product.

For a general account, the uncomfortable part is that the move arrived in the very conditions that stress the models: a selloff in which the sector's own tendency is to trade long duration, and an even widening across coupons that points to duration rather than any one prepayment view. That is exactly the risk an insurer cannot lay off easily—a whole MBS sleeve moving against the book at once, with no coupon or vintage to hide in.

The late-month move traces to Warsh's second press conference, which Patla says included ambiguous remarks that steepened the curve and lifted inflation expectations, and the coordinated currency intervention that followed added a policy variable fixed-income pricing models were not built to handle. July is therefore a preview of the general-account problem: agency MBS is increasingly a trade on the Fed chair's communication style, not just on the path of the funds rate.

Patla's 15 years of monthly buckets cover very different Federal Reserve regimes, which is reason to treat the historical averages as a weak anchor for a month when the Fed chair and the Treasury were both active. That cuts both ways: it argues against assuming July's pattern will persist, and against assuming the next 15 years will look like the last 15. The sector is now unusually dependent on the policy regime.

For insurers holding agency MBS to maturity, the spread pickup remains positive and the asset is still a balance sheet staple. For those marking to market or running a risk budget, the month is a reminder that duration protection is priced at exactly the moments it is demanded. The next test is whether a second month of curve steepening produces the same widening. If it does, the models are the lagging indicator, not the market.