What the Bond Yield Curve Shows — and What It Cannot Predict

Kabir keeps being told the curve is signalling something

Kabir has been investing long enough to have watched several of these cycles, and every so often a chart of the yield curve appears in his feed with a confident sentence attached about what it means for the year ahead. Sometimes the sentence is about a recession. Sometimes it is about what he should be holding.

He would like to know two things: what the chart actually shows, and whether the sentence attached to it is worth anything. Those turn out to have quite different answers.

The chart itself is simple enough. Plot the yield on government bonds against how long until they mature — a few months, a year, five years, thirty — and join the points. That line is the yield curve, and it answers one question precisely: what does the market currently demand to lend to the government for each length of time? Everything else attributed to it is interpretation.

Normally the line slopes upward, because lending for longer usually pays more: the lender commits money for a longer period, accepts more exposure to inflation and to rate changes, and gives up flexibility in exchange.

The three shapes

An upward slope is the usual state, where longer lending pays more. A flat curve means short and long yields have converged and the extra payment for committing for longer has disappeared. An inverted curve, where short-term yields exceed long-term ones, is unusual, and it is the shape that attracts all the attention.

The reason inversion gets discussed so much is its association with recessions in some economies, particularly the United States. The mechanism offered is plausible: short rates are high because policy is tight, and long rates sit lower because the market expects growth and inflation to weaken and therefore expects rates to be cut later.

What it genuinely tells Kabir

It tells him what lending costs now, at each maturity, which is a fact rather than an inference and is the most useful thing on the chart.

It tells him roughly what the market collectively expects, since long yields embed expectations about future short rates plus a premium for uncertainty — though the two cannot be cleanly separated, which is a real limit on how precisely those expectations can be read out of it.

And it tells him something about his own decisions, which is the part he can actually act on. A flat curve means he is being paid very little to commit money for longer, which is a live argument for staying short. A steep curve means longer lending is being rewarded. That use requires no forecast at all, which is what makes it reliable.

What the Indian curve has actually done

This page used to describe the shape and show none of it. That has changed.

Lending to the government for	Months of record	Record starts	Lowest	Typical	Highest	Most recent
1 year	362	May 1996	3.46%	6.96%	13.01%	5.68%
2 years	363	April 1996	3.88%	7.07%	13.36%	5.98%
3 years	363	April 1996	4.34%	7.16%	13.64%	6.19%
5 years	363	April 1996	4.81%	7.37%	13.79%	6.44%
7 years	363	April 1996	5.04%	7.52%	14.03%	6.63%
10 years	359	May 1996	5.11%	7.53%	13.96%	6.74%
15 years	324	May 1999	5.53%	7.65%	12.21%	6.99%
20 years	292	May 2001	5.77%	7.63%	10.13%	7.14%
30 years	196	August 2002	6.03%	7.42%	9.23%	7.32%

Two things in that table are worth pausing on. The range is enormous — lending to the government for ten years has paid anywhere from 5.1% to 14.0% depending on when you did it, which is a wider spread than most people associate with government borrowing. And the long end of the record is short: the thirty-year row rests on far fewer months than the one-year row, because those maturities were not regularly traded until much later.

The shape itself is the gap between short and long borrowing, and across 359 months from May 1996 to June 2026 it has moved a great deal.

Measured between one-year and ten-year borrowing, the typical gap was 0.82 percentage points — lending for longer normally paid more, which is the ordinary upward slope. It reached 4.10 points at its steepest, in January 1997. And it went the other way, to -1.21 points, in August 2013.

About that inversion

Now the number that decides what this page is allowed to claim.

The Indian curve was inverted — long rates below short ones — in 21 of those 359 months, or 5.8% of the time. Set a threshold of more than 0.25 points of inversion, so that a curve which is merely flat with a little noise on it does not count, and 7 months qualify.

That is not a sample. Those months are not independent either: they fall into a small number of clustered episodes, so the count of distinct events to learn from is smaller still than the count of months suggests.

So the honest position on the famous claim is unchanged by having the data, and is now unchanged for a stated reason rather than for want of evidence. We cannot tell you whether an inverted Indian curve has predicted anything, because there have been too few inversions here to test it, and because testing it would additionally need a dated history of Indian downturns that this record does not contain. A relationship established on another country's data is not evidence about this one, and the temptation to import it is exactly what the small sample should stop you doing.

What the table does establish is the part that was always the useful part: what you are being paid, today and historically, to commit money for longer.

What it cannot do

It does not predict a date. Even where inversion has preceded downturns, the gap between the signal and the event has varied enormously, and an indicator saying "something may happen within the next one to two years, or possibly not" is not actionable for somebody deciding what to hold this month.

It rests on a small sample, since the number of genuine cycles in any economy's history is small, and conclusions drawn from a handful of episodes carry far less weight than the confidence with which they are usually delivered. This is exactly the setting in which a pattern can look reliable and be coincidence, which Kabir will recognise from the overfitting article.

It travels badly between countries. The relationships most often quoted come from US data, and India has a different monetary framework, a different inflation history, different market structure and different participation. Assuming a US relationship holds here is an assumption rather than evidence.

Its shape has causes other than growth expectations — regulatory demand for long bonds, central bank operations, government borrowing patterns and foreign flows all move it, and a shape driven by supply and demand is not a forecast of anything.

And everyone can see it. Any information in the curve is public and already reflected in prices, so trading on a widely watched indicator assumes the market has failed to price a chart it stares at continuously.

What would still settle the Indian question

The rate history above closed half of this. What it does not close is the prediction claim, and it is worth being precise about what is still missing.

Two things would be needed. A dated history of Indian downturns — which quarters counted, decided in advance rather than chosen to fit — and enough distinct inversion episodes to compare against it. The first could be assembled. The second cannot be manufactured: it depends on how many times the curve has actually inverted, and in this record that is a handful.

That second constraint is not a gap money closes. It is a ceiling. Even a perfect dataset would leave the question resting on a few episodes, which is the same difficulty every country's version of this claim has and the reason it is stated far more confidently than it is known.

Using it without forecasting

The curve is most useful as a description of today's trade-offs, so ask what you are being paid to commit. If long yields barely exceed short ones, the compensation for locking money away is small, which argues for shorter deposits and shorter-duration funds. If the curve is steep, longer commitment is being rewarded.

Do not restructure a portfolio on a shape. Horizon matching — duration set against when the money is needed — is designed precisely so that the curve's shape is a detail rather than a decision.

Treat commentary with suspicion, because "the curve is signalling X" is a forecast wearing the clothes of an observation: the curve shows prices, and the signal is somebody's interpretation of them. And watch changes rather than levels, since a curve that has steepened or flattened sharply tells you expectations have moved, which is information about the present even when it forecasts nothing about the future.

What to take away

The yield curve is a clear picture of what lending costs today at different maturities, and a poor crystal ball. Its most reliable use is deciding whether you are being paid enough to commit money for longer — a question about now, which it answers directly.

Claims that it predicts recessions rest on a small number of episodes, mostly from other countries, with highly variable timing. For an Indian investor the honest position is that the data needed to test the claim here is not something we hold, and saying so is better than repeating a relationship borrowed from somewhere else.