The Polish duration trade pays over cash, except in the 2022 storm

A Polish saver who buys an open-end bond fund benchmarked to TBSP is, in effect, taking a duration position against rolling short-term deposit cash. The cash benchmark throughout this short is the NBP reference rate, used as a proxy for the best short-term deposit rates available to a retail saver in Poland (NBP bills themselves are interbank instruments and are not directly investable by an ordinary saver, but the NBP rate tracks the upper envelope of retail deposit offers closely enough that the inference goes through). The textbook justification for the trade is that the ex-ante term premium at the bond's tenor is positive on average, so holding duration earns a premium over cash. The Adrian-Crump-Moench model with the Bauer-Rudebusch-Wu bias correction agrees with that statement on the Polish curve: the sample-mean ex-ante TP is roughly +105 bp at the 10-year tenor and stays positive on the great majority of days. But the practitioner question is different. Of the 4.14 per cent compound annual return that the published TBSP delivered over the nineteen years through May 2026, how much actually came from the term-premium realisation as opposed to coupon accrual, reinvestment of cash, and the roll-down of the curve?

The seven-component identity partitions each quarter's TBSP log-return into Coupon accrual (CC), Reinvestment proceeds (RP), clean-price Roll-down (RD), Expected-rate change (ERC), Term-premium change (TPC), Convexity (Cnvx) and a small closure Residual (Res). For the picture below I group the first three into what I call CarryRoll, and the next two into CurveMove. Both names are my own informal labels for this short and have no status beyond it. They are not benchmarks, not maintained as part of any index methodology, and not published anywhere as an index family. Their only purpose is to compress the seven-component identity into two readable lines that map cleanly onto the practitioner question above. Convexity and Res are second-order. Cumulating each component across the eighty quarter-end snapshots of TBSP yields a clean answer.

The cumulative attribution is uncompromising. Carry-and-roll (CC + RP + RD) contributes +8,010 bp, or +413 bp per year on average, which by itself almost matches the +7,866 bp delivered by the published TBSP. The curve-move bloc (ERC + TPC) nets to −337 bp over the full sample, or −17 bp per year, the small negative reflecting the 2022 NBP hiking-cycle drawdown that did not fully reverse. Convexity contributes a quiet +279 bp. The closure residual is +20 bp and the tracking gap against the published index is −106 bp, both small enough that the seven-component identity is doing its work as an attribution rather than as a Procrustean bed.

On the full sample, the formal inference is uninformative. The realised TPC point estimate of −3.1 bp per quarter has HAC p = 0.83 and a 95 per cent bootstrap CI on the cumulative TPC of [−2,534, +1,776] bp. The TBSP-versus-NBP-deposit excess (+17.6 bp per quarter, +1,354 bp cumulative) has HAC p = 0.55. Both bracket zero. But the full-sample picture is dominated by a single four-quarter regime episode: 2021-Q4 to 2022-Q3 contributed −962 bp of cumulative TPC, four times the entire sample's −245 bp. Conditioning out those four storm quarters, the TBSP-versus-rolling-NBP-deposit excess is +46.5 bp per quarter with HAC p = 0.003 and a cumulative bootstrap CI of [+1,217, +5,534] bp on the 74-quarter calm sub-sample. The duration trade compensated the saver over deposit-style cash on 74 of the 78 quarter-ends in the sample. The four quarters of 2022 produced a −21.7 per cent peak-to-trough TBSP drawdown that the post-2023 recovery only partially undid.