Function priceRelative

Price Relative — the ratio of column to benchmark, bar by bar:

${output} = column / benchmark
Copy

Appends one column, with no period and no warm-up: it is the only study in the package that reads a single row, so every row emits (bar 0 included) wherever both inputs exist. A rising line means column is outperforming, falling means the benchmark is — the level is arbitrary (it carries the two instruments' price units), which is why the line is read for its slope and why performanceIndex exists to normalise it against a starting point.

"Relative Strength" here is not rsi

ChartIQ lists this under two names — Price Relative and Relative Strength (comparative) — and they are one implementation, which is what ships here under the unambiguous name. Neither is Wilder's Relative Strength Index: rsi is a single-series momentum oscillator bounded 0..100, this is an unbounded two-series ratio, and the collision of names is the single most common confusion in the corpus. If you want RSI, call rsi.

The comparison series is a COLUMN, not a second TimeSeries

Join the benchmark in first and name its column:

const wide = TimeSeries.joinMany([bars, spy.rename({ close: 'spy' })], {
  type: 'inner',
});
priceRelative(wide, { column: 'close', benchmark: 'spy' });
Copy

See correlation's docstring for the reasoning (aligning two bar clocks is align/joinMany's job, not a study's). A missing benchmark column throws.

Edges

• No TA-Lib function, so the oracle is a pandas replication, with the inverted ratio (benchmark / column) measured as the separation — the one substitution that leaves the curve's shape recognisable and every value wrong.
• A zero benchmark reads undefined, and the guard is LIVE. Unlike the flat-window cases in correlation and beta, the numerator here is not forced to zero with the denominator, so 5 / 0 is Infinity — and withColumn throws on a non-finite value rather than recording it as a gap. The guard sits at the study's output, which is what makes it observable (the #703 rule: a guard upstream of a window kernel is absorbed and dies; one at the output is load-bearing). Unreachable on prices, reachable when benchmark is another study's output that crosses zero, and unit-tested there. A negative benchmark still produces a number — === 0, not <= 0, the percentChangeValues rule, because a ratio against a negative level is defined if unusual and clamping it would invent a rule.
• Scale-EQUIVARIANT, not invariant. Multiplying column by k multiplies the reading by k; multiplying benchmark by k divides it by k. Both are pinned as property tests, as identities rather than as "the numbers changed" — this is the one study in the family that is not scale-invariant, and a test that asserted invariance would have passed on a study that had normalised the ratio away.
• A missing cell in either column blanks that row and nothing else — a window kernel's recovery rule with a window of one.
• benchmark must differ from column: the ratio would be 1 everywhere.