🎯 SKM Egg Products — egg-spread cycle model

The spread

ln(US egg price × INR/USD) − ln(Namakkal egg price). The output side is what SKM sells powder for abroad; the input side is what it pays for shell eggs at home.

The signal

The 12-month change in that spread, placed in a quartile of its own history as known at the time.

Why a lag

The same-month correlation between US egg prices and the share price is about +0.03 — nothing. It rises with lag to a plateau of +0.24 to +0.29 across 8 to 13 months, the time it takes for export contracts to reprice and for shipments and earnings to follow.

Which leg matters

Tested head to head on forward 12-month return relative to the market, the Namakkal input cost carries a coefficient of −1.38 (t = −3.19) against the US price’s +0.62 (t = +2.10). The two egg prices are near-uncorrelated (r = −0.13), so they carry independent information.

How the band is built

The band takes its centre from what actually happened after past calls carrying the same label, and its width from the full return history — a small pool can estimate a median but not a tail. Raw quartile bands measured 32% coverage against a nominal 60%, so a 1.25× dispersion correction is applied.

Liquidity cut

Nothing before September 2014 is used. Before that SKM traded ₹0.01–0.75 crore a month, so the printed close is a stale quote rather than a market price.

How to read a zone

Each monthly reading describes the 12 months that follow it, so the way to act on it is to hold that call for twelve months rather than to trade the zone as it flips. The label lasts a median of one month; the claim it makes lasts twelve.

Tested against random signals

The same machinery was run with the egg spread replaced by 399 unrelated commodity prices, holding the stock, the horizon and the benchmark fixed. None of the 399 matched it — on the zone spread or on realised return. That test failed for other candidates we did not publish.

Nothing here has hindsight

Every historical call uses an expanding window: the quartile cuts at month t use only months up to t, and the outcome pool only months up to t−12, whose result was actually known by then. Bands pool on the label a month was given at the time, never by re-sorting history under today's cuts. The record above is what this model would have published then.

Where it breaks

• The sample is tiny. Completed calls over overlapping 12-month windows are roughly six independent episodes. Every number here has wide error bars.
• Followed month to month rather than held for twelve, the bullish side is no better than picking an unrelated commodity at random. The holding period is doing the work.
• It understates upside in bullish zones — the misses are almost all above the band, not below.
• The 1.25× width correction was fitted on this sample. Its sign is well founded; its exact value is not.
• Feed costs and SKM’s own realisations are missing. The Namakkal price is itself downstream of maize and soymeal, and the US shell-egg price is not what SKM actually gets for powder.
• It is a microcap. Company-specific events routinely dwarf any commodity signal.