15.9 Playbook 9: crude oil into an Indian stock, the EIA-Wednesday rule (BPCL, 4–6 November 2026)

15.9.1Prerequisites and what this playbook teaches#

Read first (essential) M10 · 10.8 (crude oil, the EIA report, India's import dependence and the oil marketing companies), M12 · 12.1–12.2 (event-study design and the abnormal-return $t$-test; the same market-model primer is in M06A · 6A.2), M03 · 3.1.1 (how many trades it takes to detect an edge) and 3.9 (position sizing and gap risk), M04A · 4A.4 and 4A.7 (gaps and ATR stops), M01 · 1.2 (the pre-open and closing auctions), M01 · 1.7–1.8 (costs and tax).

For depth: M12 · 12.3 (cross-sectional and clustered tests, for the version of this study over all three oil marketing companies).

What it teaches How a commodity shock that happens while the Indian market is shut reaches an Indian stock; how to tell how much of a stock's move is the shock and how much is the market; how to size a position when the main risk is a gap through the stop rather than a slow move to it; and what an honest base rate looks like for a rule that sounds obvious.

This is an educational reconstruction of a rule applied to past data, made on paper. It is not a recommendation to buy or sell BPCL or any other security.

The thesis in one paragraph Bharat Petroleum (BPCL), Hindustan Petroleum (HPCL) and Indian Oil (IOC) are oil marketing companies (OMCs): they buy crude, refine it and sell fuel at pump prices that the government influences and that move slowly. A fall in crude widens their marketing margin, so OMC shares should rise when crude falls (M10 · 10.8.5). The US Energy Information Administration (EIA) publishes its Weekly Petroleum Status Report on Wednesdays at 10:30 ET, which is 21:00 IST in US winter time (from 2 Nov 2026), after the NSE cash market has closed. So a large Wednesday move in crude reaches Indian OMC shares through Thursday's opening auction. The trade tests one question: once the opening auction has priced the move, is there anything left to earn?

15.9.2The setup as testable rules#

All prices are Yahoo Finance daily bars: BZ=F (ICE Brent front month) and BPCL.NS. Every rule uses only data available before the order is placed.

1. Event set. Every Wednesday on which Brent futures trade. These are EIA release days. A US holiday sometimes moves the release to Thursday; the rule ignores those shifts, which slightly misdates a few events.
2. Signal. The Brent close-to-close log return on the Wednesday, $r_W$, satisfies $r_W \le -2\hat\sigma_{60}$, where $\hat\sigma_{60}$ is the sample standard deviation of the 60 daily Brent log returns ending on the Tuesday. A fall of two "normal days" is the threshold. The rule does not ask why crude fell.
3. Instrument. BPCL, the most liquid of the three OMCs in the cash market. Long only, so cash delivery is enough (M14 · 14.17 covers why shorting cash shares overnight needs SLB).
4. Entry. Buy at Thursday's opening auction price (the next NSE session after the Wednesday). The NSE open is a single-price call auction that ends at a random time between 09:08 and 09:10 (M01 · 1.2, as of 6 Nov 2026), so a limit order placed in the pre-open at a generous price fills at the open. Stop-loss orders are not accepted in the pre-open; the stop is entered at 09:15.
5. Stop. Entry minus $1.5\times\text{ATR}_{14}$, where $\text{ATR}_{14}$ is the simple mean of the last 14 true ranges up to Wednesday's close (M04A · 4A.7). A stop-market order: if the session low touches the stop, the fill is the stop; if the next session opens below it, the fill is the open.
6. Exit. Sell in the closing auction of the session after entry (Friday, when the entry is on a Thursday). Two sessions is the window in which the thesis says any drift should appear, and it keeps the position flat over the weekend, when geopolitical news arrives with no market open to trade it.
7. Invalidation. If Brent recovers more than half of the Wednesday fall by Thursday's US close, exit at Friday's open instead of the close.
8. Size. Fixed-fractional with a gap cap, below.

Gap-adjusted sizing Let $E_q = \text{₹}10{,}00{,}000$, $f = 1\%$, $d$ the stop distance per share and $P$ the entry price. Let $g_{99}$ be the 99th percentile of BPCL's adverse overnight gaps, $1 - O_t/C_{t-1}$, over the five years to the signal date. A gap of $g_{99}$ that opens through a stop just after a close at the stop costs about $d + g_{99}P$ per share. The rule caps that loss at $1.5fE_q$:

The first term is (3.23) of M03 with lot size 1. The second term binds when the gap tail is wide relative to the stop, which is the normal case for a stock that has just gapped on overnight news.

15.9.3The scan and the episode#

Data Yahoo daily bars for BZ=F, CL=F, BPCL.NS, HINDPETRO.NS, IOC.NS, ^NSEI and INR=X, from 2010 to 6 Nov 2026, downloaded on 6 Nov 2026. Yahoo's equity prices are adjusted for splits and bonus issues but not for dividends; R-multiples do not depend on the price scale. Futures lot sizes are from the NSE F&O bhavcopy of 5 Nov 2026 (NewBrdLotQty). Weekly crude stocks are from the EIA (series WCESTUS1).

The 2026 scan In 2026 the rule fired three times: 1 Apr ($r_W = -15.7\%$ in logs, $z = -4.17$), 8 Apr ($-14.3\%$, $z = -3.24$; the two-week ceasefire of M12 · 12.8, Case 6) and 4 Nov. The 4 Nov signal is the most recent one before this module's date, so it is the trade. The next-largest 2026 Wednesday fall relative to $\hat\sigma_{60}$ was 6 May, at $z = -1.57$.

The week of the signal

Date	Crude news	Brent close ($)	BPCL close (₹)
Fri 30 Oct	—	117.64	284.90
Sun 1 Nov	OPEC+ (seven countries) keeps December targets at September levels (OPEC press release, 1 Nov 2026)	—	—
Mon 2 Nov	Brent opens $111.84 ($-4.9\%$), closes $-10.9\%$	104.77	293.05
Tue 3 Nov	US midterm elections	103.60	296.90
Wed 4 Nov	EIA: commercial crude stocks (excluding the SPR) fell 0.865 mb in the week to 30 Oct, from 410.625 to 409.760 mb (EIA series WCESTUS1). Market consensus was reportedly for a larger draw, about 1.7 mb (not verified against a primary source). Democrats win both houses; peace hopes (M12 · 12.8, Case 7)	94.62	304.70

On 4 Nov $r_W = -9.07\%$ (log; $-8.67\%$ simple) against $\hat\sigma_{60} = 2.63\%$, so $z = -3.44$. WTI (CL=F) fell from $96.09 to $88.12, $-8.29\%$. The rule fires.

Read the table before trusting the thesis. The EIA print was a smaller draw than the reported consensus, which is mildly bearish for crude, but nothing like a 9% day. The day's news was the midterm result and what it implied for an Iran settlement. An OPEC+ decision two days earlier, a hold of output targets, was followed by an 11% fall in Brent, the opposite of what a hold usually implies. On daily bars a scheduled release cannot be separated from the unscheduled news of the same day (M12 · 12.1, overlapping windows). The rule does not need to know the cause. The post-mortem does.

15.9.4The plan before entry#

Written on the evening of Wednesday 4 Nov 2026, after Brent's settlement.

Thesis Brent fell 8.7%. The OMC marketing margin widens; BPCL should open higher. The trade bets that the open under-reacts and that BPCL drifts higher over two sessions.

Trigger Buy at Thursday's opening auction. No price filter: the rule takes the open whatever the gap.

Stop $\text{ATR}_{14} = \text{₹}8.457$ on 4 Nov. The entry is unknown until the auction, so the stop is fixed as a distance, $d = 1.5 \times 8.457 = \text{₹}12.686$, and the price is computed at 09:15.

Size BPCL's adverse 99th-percentile overnight gap over 1,235 sessions to 4 Nov 2026 is $g_{99} = 2.67\%$; the worst was $-6.24\%$. The gap term depends on the entry price, so $Q$ is computed from the indicative pre-open price and fixed before order entry closes. At the actual open of ₹320.00, (15.9.1) gives

so $Q = 706$ shares, notional ₹2,25,920 (22.6% of capital). The risk at the stop, $1R$, is $706\times12.686 = \text{₹}8{,}956$. A 99th-percentile gap through the stop would cost ₹14,980 ($1.67R$); a repeat of the worst gap in five years, ₹23,063 ($2.58R$).

Why not futures One BPCL futures lot is 1,975 shares (NSE F&O bhavcopy, 5 Nov 2026). At $d = \text{₹}12.686$ one lot risks ₹25,054 at the stop, 2.5% of capital. The smallest futures position is too large for this account at this stop, so the trade uses cash delivery.

Costs (M01 · 1.7, statutory and exchange charges as of 6 Nov 2026) Delivery: STT 0.1% on each side, exchange 0.00307%, SEBI ₹10 per crore, stamp 0.015% on the buy, GST 18% on exchange and SEBI charges, a depository (DP) charge of ₹15.34 on the sale, and brokerage set to zero as an illustrative parameter (delivery brokerage varies by broker; add yours). With $\alpha = 0.0011874$, $\beta = 0.0010374$ and $K = \text{₹}15.34$, the break-even move of (1.4) is ₹0.73 a share, 0.23% of the price, or $0.058R$.

Expected R from the record available on 4 Nov Applying the same rules to every earlier Wednesday from 2011 gives 35 signals. Net of costs the mean was $-0.053R$, standard deviation $0.66R$, 51.4% winners, average winner $+0.42R$, average loser $-0.55R$. The break-even win rate at those averages is $0.55/(0.42+0.55) = 57\%$. The $t$-statistic of the mean is $-0.48$. Before costs the mean was $-0.002R$: the rule had no gross edge and costs of about $0.05R$ a trade made it negative.

A trader following M14 · 14.10.2 would not trade a setup with these statistics. It is taken here, on paper, as a pre-registered test of the transmission thesis, and the size is the full rule size so that the costs and the R accounting are real. Expected result: $-0.05R$, about $-\text{₹}480$.

15.9.5Day-by-day management#

Date (IST)	Market	BPCL O / H / L / C (₹)	Action and reasoning	Open P&L (₹, gross)
Wed 4 Nov, 21:00	EIA report; Brent settles $94.62	296.50 / 304.70 / 295.50 / 304.70	Signal confirmed after Brent's close. Pre-open buy order for 706 shares with a limit well above the expected open.	—
Thu 5 Nov, 09:08–09:10	Nifty opens higher; closes 24,536.60 (+2.4%)	320.00 / 322.00 / 312.30 / 317.10	Filled 706 at the auction price ₹320.00, a gap of +5.02% over Wednesday's close. At 09:15 a stop-loss market order is entered at ₹307.30 (the computed ₹307.31 rounded down to the 5-paise tick). The low of ₹312.30 stays ₹5 above it.	−2,047
Thu 5 Nov, evening	Brent closes $98.29 (+3.9%)	—	Invalidation check: half the Wednesday fall is $94.62 + 0.5\times8.98 = \$99.11$. Brent is below it, so the thesis stands and the exit stays at Friday's close.	—
Fri 6 Nov	Brent $96.88	312.90 / 319.90 / 308.60 / 319.90	Opened 1.3% lower. The low of ₹308.60 came within ₹1.30 of the stop. Sold 706 in the closing auction at the official close, ₹319.90.	−71

Nothing in the log required judgement. That is the point of writing the plan the evening before: the only decision on Friday morning, with the price ₹1.30 above the stop, was to leave the stop where it was (M04A · 4A.10.2).

15.9.6Exit and result#

Item	₹
Buy value $706\times320.00$	2,25,920.00
Sell value $706\times319.90$	2,25,849.40
Gross P&L	−70.60
STT (0.1% × both sides)	451.77
Exchange charge	13.87
SEBI fee	0.45
Stamp duty	33.89
GST	2.58
DP charge	15.34
Total costs	517.90
Net P&L	−588.50

In R: gross $-70.60/8{,}956 = -0.008R$; net $-0.066R$. Costs were eight times the gross loss.

Tax A delivery trade held overnight with STT paid is a short-term capital transaction (M01 · 1.8; M00 · 0.7 for the rules a beginner needs). STT is not deductible in computing capital gains (M01 · 1.8); the other charges (₹66.13 here) are normally treated as costs of acquisition or transfer. The short-term capital loss for tax is therefore about $70.60 + 66.13 = \text{₹}136.73$, not the ₹588.50 economic loss. Set off against short-term gains taxed at 20%, it saves about ₹27 of tax. The after-tax result is about $-\text{₹}561$. Someone who trades this often may be treated as carrying on a business, with the income taxed at slab rates instead; that is a question of facts for a chartered accountant, not a choice.

15.9.7Post-mortem#

What the event study says happened on 5 Nov Take $\tau = 0$ as 5 Nov, the first NSE session after the Wednesday release (M12 · 12.1, time zones). Estimate the market model (12.1) against the Nifty over $\tau\in[-250,-11]$ ($L_1 = 240$) and test with (12.7) and (12.9).

Stock	$\hat\beta$	$\hat\sigma$	Return on 5 Nov (log)	$AR_0$	$t_0$	$CAR[0,1]$	$t$
BPCL	1.38	1.41%	+3.99%	+0.71%	0.49	−1.16%	−0.56
HPCL	1.56	1.79%	+3.67%	−0.00%	−0.00	−1.23%	−0.47
IOC	1.12	1.34%	+1.72%	−0.90%	−0.66	−1.88%	−0.97

The Nifty rose 2.38% (log) on 5 Nov. With a beta of 1.38, BPCL's "normal" return was 3.28%, and its excess over that was 0.71%, well inside noise. None of the three OMCs had a significant abnormal return on the day or over two days. The rupee (INR=X) moved from 94.75 to 94.52 per dollar on 5 Nov, a 0.25% appreciation, with a constant-mean $t$ of $-0.57$.

The table has a trap in it, which M12 · 12.1 warns about. Crude fell and the Nifty rose on the same news. A market model that uses the Nifty as the benchmark assigns the common part of the crude effect to "the market", so the OMC abnormal return measures only the OMC-specific part. The honest reading is narrower than "crude did not matter": on 5 Nov the OMCs did no better than their market betas implied, so there was no OMC-specific under-reaction to buy.

What the history says Over all 811 Wednesdays from 2011 to 2026, regress BPCL's next-session returns on Brent's Wednesday log return:

BPCL return	Slope on $r_W$	$t$
Overnight gap (Wednesday close to Thursday open)	−0.060	−4.24
Thursday open to close	+0.015	0.52
Thursday open to Friday close (the trade)	−0.039	−0.91

The transmission is real and has the expected sign: a 1% fall in Brent adds about 0.06% to BPCL's Thursday open. It is small, and it is all in the opening auction. The slope from the open onward is indistinguishable from zero. The thesis needed drift after the open, and there is none. Part of the full transmission is not in the gap at all: Brent's Wednesday return includes Asian and European hours that overlap Wednesday's NSE session, so some of the effect is already in Wednesday's close.

The rule's base rate, including this trade $N = 36$ signals from 2011 to 2026 (about 2.3 a year). Gross: mean $-0.002R$, standard deviation $0.65R$, 58% winners. Net of costs: mean $-0.054R$, 50% winners, $t = -0.50$. Exits: 31 at the time stop, four at the stop, one through a gap (17 Apr 2024 signal, $-1.005R$). The 4 Nov trade, at $-0.066R$ net, is a typical draw from this distribution.

How many trades would settle it Suppose the rule had a true net edge of $E = +0.10R$ with $s = 0.65R$. By M03 (3.2a) it would take $(2\times0.65/0.10)^2 = 169$ trades for the expected $t$-statistic to reach 2, and by (3.2b) 341 trades for 80% power. At 2.3 signals a year that is 150 years. A rule that fires this rarely cannot be validated on its own signals; it can only be justified by a mechanism that is tested on a larger event set, as the all-Wednesday regression does.

The selection trap The mirror rule (short BPCL futures after a Brent rise of $2\hat\sigma_{60}$ or more) had 23 signals, a gross mean of $+0.13R$ and a $t$ of about 1.06 before costs. Switching to it after seeing this table would be the specification search that M03 · 3.1.1 shows inflates false positives.

What went right The plan was written before the open and followed. Gap-adjusted sizing kept a 99th-percentile gap within $1.5\%$ of capital. The futures alternative was rejected on arithmetic, not taste.

What a better version would need (i) A surprise measure: actual inventory change against consensus, which needs historical consensus data that are not freely available, or against a five-year seasonal average from the EIA's own data (M10 · 10.8.2). (ii) Intraday data, so that the 21:00 IST release can be separated from the rest of the day's news. (iii) Fewer costs per unit of edge: a two-day hold that pays 0.2% of notional in STT needs a large expected move to clear it.

15.9.8Do it in the simulator#

The simulator is end-of-day: cash equities fill against Yahoo daily bars, a limit or stop order becomes eligible from the next session's open, a stop that the market gaps through fills at the open, and a market order fills at that day's close (or at the open if the account's market-fill setting is "open"). It has no stock futures and does not allow overnight short sales of cash shares, so the long rule can be run exactly and the mirror short rule of 15.9.7 cannot be run at all.

Open Simulator → New order, with a ₹10,00,000 paper account, and run the rule on the next signal it gives after 6 Nov 2026. Until one comes, use it to check your process on the most recent completed signal.

1. On a Wednesday evening, compute $\hat\sigma_{60}$ from the 60 Brent log returns to Tuesday and check whether that day's $r_W \le -2\hat\sigma_{60}$. If not, there is no trade; record that in the journal (a skipped signal and a non-signal are both data).
2. On a signal, compute BPCL's $\text{ATR}_{14}$ to Wednesday's close and $g_{99}$ over the previous five years. Size with (15.9.1) using Wednesday's close plus the expected gap as the entry estimate, and state which term binds.
3. Enter a buy limit order that evening with the limit 10% above Wednesday's close, so that it fills at Thursday's open whatever the gap (on 5 Nov 2026 a 3% limit would have missed the +5.02% open, and the rule says to take the open); attach a stop-loss bracket at entry minus $1.5\,\text{ATR}_{14}$ and write the thesis and the invalidation level for Brent in the order form. The Preview shows the costs, the risk amount and the percentage of equity at risk; check the risk against ₹10,000 before submitting.
4. On Friday, close the position with a market order (default fill: the close). If the stop was gapped through on Friday's open, note that the simulator filled it at the open, as 15.9.2 assumed.

See it live: BPCL's current price and futures OI by expiry at https://oicharts.com/stocks/BPCL . Brent is not on that site; take it from the source you use for the rule. 5. Record the net R from the journal and place it against the base-rate table in 15.9.7.