Basis Swaps and Tenor Basis
Tenor basis is best understood as a priced package of cash-flow conventions and risk factors, not as a single market quote. The headline level is useful for communication, but a quantitative implementation must reproduce contractual cash flows, identify the curve and fixing data used at every valuation date, and explain why the hedge can deviate from the economic thesis. That discipline turns a plausible relative-value story into a trade that can survive daily marks, collateral calls, and stressed exits.
Economic object and valuation
In this context, tenor basis means the spread exchanged when two floating benchmarks or reset frequencies are swapped against one another. A desk should begin with a trade-definition object: effective and termination dates, notional schedule, pay/receive direction, calendars, day counts, payment dates, and collateral currency. The relevant pricing object is usually a multi-curve framework; it is not interchangeable with single-curve parity. The distinction matters most around policy shocks, month-end funding, and periods where liquidity is rationed.
A useful generic decomposition is:
market quote = expected cash flows + risk premia
+ liquidity/funding effects + convention effects
PV = sum(expected contractual cash flow * collateral discount factor)
The first line is a research decomposition, not an identity that can be estimated without error. It prevents a common mistake: interpreting every movement as a change in the macro expectation. The operational details include separate discount and forwarding curves, reset dates, fixing lags, collateral terms, and quote-side conventions. Store each of those inputs in the valuation record. A model that gives the right price with undocumented defaults cannot be independently reconciled.
| Layer | Question for the quant | Typical failure |
|---|---|---|
| Contract | What cash flow is legally exchanged? | Applying a generic schedule |
| Market data | Which executable quote enters calibration? | Using stale composite mids |
| Curve | Which discount and projection assumptions apply? | Mixing curve dates |
| Risk | Which market quote is bumped? | Reporting an ambiguous DV01 |
| Execution | Can the hedge trade at the modeled price? | Ignoring bid/ask and capacity |
A transparent Python check
The following intentionally small calculation is a control, not a production pricer. It makes discounting and sensitivity definitions explicit before a framework hides them:
import numpy as np
def pv(cashflows, discount_factors):
return float(np.dot(cashflows, discount_factors))
def bumped_sensitivity(cashflows, dfs, bump=1e-4):
up = np.asarray(dfs) * np.exp(-bump * np.arange(len(dfs)))
down = np.asarray(dfs) * np.exp(bump * np.arange(len(dfs)))
return (pv(cashflows, down) - pv(cashflows, up)) / 2.0
This is deliberately stylized: dates should replace array indices, and a market-quote bump should trigger a curve rebuild. Still, an independent implementation like this catches sign errors and makes the risk report auditable. For the underlying swap plumbing, see interest-rate swaps. Rate products also require a clear separation between valuation under collateral and any separate funding valuation adjustment.
Signals, relative value, and hedging
Research should combine forward-basis curves, futures-implied forwards, FRA/OIS measures, bank funding proxies, and quarter-end balance-sheet effects. Use only information available at the decision timestamp, preserve source timestamps, and distinguish indicative from executable prices. Fit signals to residuals after carry and known seasonal effects rather than to raw levels. A level can look predictive simply because the curve rolls mechanically through a calendar event.
One implementable expression is to receive a rich short-tenor spread versus pay a hedged longer-tenor spread when forward basis curvature dislocates. First neutralize the intended dominant risk with key-rate or factor hedges, then retain the factor you mean to own. A scalar duration hedge is rarely adequate: curve twists, basis, option convexity, and liquidity can dominate a headline DV01. For inflation-linked applications, TIPS breakeven construction provides the cash-market complement; for funding across currencies, compare the mechanics in cross-currency basis swaps.
| P&L driver | Measurement | Hedge or control |
|---|---|---|
| Curve move | quote-bucket repricing | liquid futures or swaps |
| Carry | roll-forward on unchanged curve | include in signal hurdle |
| Basis/technical | residual versus peer instruments | relative-value hedge |
| Volatility | scenario revaluation | convexity budget |
| Liquidity | conservative liquidation cost | size and concentration limit |
Backtest at a realizable frequency. Use bid/ask, market holidays, contract rolls, and fixing cutoffs. Cross-validation cannot rescue a signal whose prices are revised after the fact. Attribute live P&L into curve, carry, fixing, spread, and execution components; a persistent residual is evidence of a missing risk factor, not a reason to widen a catch-all “model error” bucket.
Stress testing and governance
Design scenarios around the mechanism that invalidates the thesis: abrupt policy repricing, a correlated liquidity withdrawal, a fixing surprise, or a spread gap while hedges become one-sided. Revalue full cash flows rather than scaling a normal-volatility VaR. Include calling the quoted spread a pure credit signal, ignoring initial exchange conventions, and hedging only aggregate DV01. Counterparty exposure, margin timing, and settlement fail risk belong in the same pre-trade review as forecast accuracy.
Where the payoff depends on volatility or correlation, the static intuition from variance-swap replication is valuable but incomplete. Option-strip liquidity, jump exposure, and model extrapolation become first-order. Related dispersion implementation is covered in volatility arbitrage and dispersion; equity-financing assumptions should be reconciled with dividend futures and equity swaps.
Key takeaways
- Treat tenor basis as contractual cash flows plus curves, conventions, and liquidity—not a standalone forecast.
- Build signals from timestamped, executable data and remove mechanical carry before judging alpha.
- Hedge quote-bucket and factor exposures, then stress the residual risks that a simple DV01 cannot capture.
- Reconcile daily P&L into economic components so model errors become observable and actionable.