Bottom line

The rupee’s strength has a date on it. The fall in USD/INR is real and still gathering pace, but it is being powered by a dollar inflow that stops this Friday. Once it stops, the support underneath the rupee thins out sharply — and oil at $97 with foreign investors still selling is waiting on the other side.

Add to that a market already sitting on a well-defended floor with daily movement shrunk to a fraction of normal, and the balance of risk for the rest of September tilts the other way.

Exporters: stop waiting. Sell into strength from 94.45 upward this week. Importers: this is probably your last cheap week. Take it in stages rather than holding out for a level the flow may no longer deliver after Friday.

What actually moved the rupee

In June the RBI opened a special dollar–rupee swap facility for fresh FCNR(B) deposits. It absorbed the banks’ full hedging cost, removed the interest-rate ceiling and exempted the deposits from CRR and SLR. Banks could therefore offer non-resident depositors 6% to 7% on dollar fixed deposits with no rupee risk — against 3.5% to 4% before.

The response overwhelmed the design. By 13 August the facility had drawn $52.3 billion through FCNR(B) alone, close to $57 billion across all three channels. For scale, the comparable 2013 window raised $27 billion — and charged banks 3.5% for the hedge, where this one charged nothing.

On 14 August the RBI responded by pulling both deadlines forward: deposit-raising from 30 September to 31 August, and the swap execution window from 16 October to Friday 11 September.

This matters more than any indicator on the page, for one reason: it is a flow with an end date, not a change in India’s fundamentals. Nothing about the trade deficit, the oil bill or foreign investor appetite has improved. When the flow stops, the pressures it has been masking are still there.

What is documented: the scheme, the $52.3 billion figure, the deadline change and the 11 September cut-off. What is inferred: that this is what drove the move. No day-by-day conversion data is published, so we are matching a mechanism to a timing rather than measuring a flow directly. We would rather say that plainly than present it as fact.

Where the rate is

The rate is at its lowest in fifty trading days and sits close to the bottom of that range — yet it is still three-quarters of the way up its one-year range. Cheap by recent standards, still expensive by yearly standards. That is the tension every hedging decision this week has to sit inside.

Where we expect the rate to trade · 8–11 September

Closing prices and intraday extremes behave differently, so we publish both. Use the closing band to judge the week; use the intraday reach to place orders. This is a four-day week, Tuesday to Friday, and both bands are measured over four sessions.

Three ways the week can go

Weights are estimates of historical frequency, not forecasts. Note that no outcome carries even half the weight, and note the shape of the week the three imply together: the rupee-positive case is front-loaded and the dollar-positive case is back-loaded. Execution timing inside the week matters more than usual.

The four levels that change the plan

If you are an exporter — receiving dollars

Sell in tranches as the market rises. Do not park the whole order at one price: the market has not closed above 94.97 in five sessions, and 94.51 is where bounces have historically stalled.

If you are an importer — paying dollars

You have had a good fortnight. Treat it as finite: the flow behind it ends Friday, and the level you are buying is 95 paise better than two weeks ago.

Today’s instruction

Three sets of evidence feed this week’s view: what the market actually did, what the outside drivers are doing, and what the forward market is charging. They do not fully agree, and we say where they disagree rather than smoothing it over.

What the market did — five facts

1. The fall is unusually large and unusually fast. USD/INR has closed lower in seven of the last eight sessions, losing 95.5 paise against our last published anchor and 2.22% over thirty sessions. It now sits 7% above the bottom of its fifty-day range, having spent August in the upper half of it.

2. Oil rose sharply and the rupee ignored it. Brent closed at $97.14, up 7.38% on the week and 16.30% on the month. The wider energy complex moved with it — European gas up 7.23%, gas oil up 5.09%, heating oil up 4.27%. India imports more than 80% of its crude, so this is normally the single fastest input into the rate, and it normally works against the rupee. This week it did not register at all. That is the clearest evidence that something non-market was setting the price.

3. Foreign investors sold, and domestic institutions absorbed it entirely. Over the five sessions to 4 September, foreign institutions were net sellers of ₹5,612 crore while domestic institutions bought ₹23,156 crore. Foreign investors also held a net short position of 235,838 index futures contracts on Friday. Ordinarily a week of foreign selling on that scale pressures the rupee. It did not.

4. Indian shares stayed weak. The Nifty fell 1.09% on the week, is down 3.06% on the month and 8.85% for the year. The Sensex is down 10.42% year to date. This is the one domestic signal that has been working against the rupee for months, and it has not improved.

5. Global markets were calm. US equities finished the week slightly higher, the main US fear gauge fell 2.61%, and gold and silver both eased. Nothing in global sentiment currently argues for a weaker rupee. This is a mild tailwind, not a driver.

What the price action says, without the jargon

What the outside world is doing

Two conflicts we are not going to hide

The trend and the stretch disagree, and there is no precedent for the combination. Our strength measure says this is one of the most decisive downtrends in the dataset. Our stretch measure says the market is close to the most extended it has ever been. Searching 669 candidate days, the closest historical match to today’s conditions scored 53.9 against a 60 threshold — meaning there is no genuine precedent. Strong trends can stay stretched far longer than mean-reversion statistics suggest. Read the historical numbers in Appendix B with that caveat attached.

Our own models disagree with each other. The historical evidence leans mildly higher — comparable conditions ended higher 58% of the time within a week. The market structure leans lower: five structural factors support further downside and none support upside. When the two disagree, the pair usually spends time going nowhere before resolving, which is a poor environment for one large hedge decision and a good one for staged execution.

Putting the three together

Price behaviour: a decisive downtrend that is simultaneously more stretched than almost anything in our records, moving less each day as it goes.

Outside drivers: one very large supportive flow with a Friday expiry, against oil, foreign selling and weak equities all pointing the other way. Remove the flow and the balance inverts.

Forward market: the cost of deferring rose beyond three months and fell inside a month — the market is charging more to wait longer, not less.

Together: mildly bearish this week, with the risk shifting decisively the other way from Friday. Confidence on direction is medium. Confidence on the range is medium — lower than the recent calm alone would justify, because two large events and a flow deadline all land in the same 48 hours.

The one date we could not fully reconcile

Two published accounts of the RBI’s 14 August announcement differ on whether the swap execution deadline moved to 11 September or stayed at 16 October. The weight of reporting, including the RBI’s own framing, supports 11 September, and we have written the note on that basis. If it is in fact 16 October, the flow runs five weeks longer and the bounce risk moves out with it. Please confirm against your own bank before acting on the timing. This is the only load-bearing date in the note we cannot triple-source.

Just beyond this week

The honest read on this week’s calendar

Monday was closed and Tuesday is empty. Everything that matters is compressed into Thursday and Friday, and Friday alone carries three separate catalysts pointing the same way. If you have exposure settling in the first half of September, plan around Thursday and Friday — not around the average of the week.

A working table. Find where the rate is trading, then read across. Use it during the week, not once.

Adjusting for when your money moves

The risk of waiting, both ways

Exporters: the market has not closed above 94.97 in five sessions. A single unfilled order parked at 95.05 is the mistake to avoid — but so is assuming the flow that pushed the rate here will persist. Your risk is that Friday passes, the rate recovers, and you sold nothing on the way up because you were waiting for a level you would have reached anyway.

Importers: you have had the best fortnight since the spring, and the mechanism that gave it to you has a published end date. In comparable historical situations the rate reached the favourable zone below only 40% of the time within five sessions, against an 85% chance of reaching the zone above. Take what is available in stages rather than holding out for 94.10.

Booking a rate for a future date costs more than today’s rate. This week the rate itself fell more than a rupee — but the extra cost of booking ahead moved in two directions at once, which is the more interesting fact.

Desk forward calculator, standard month-wise rates, spot value date 9 September 2026. “Change on week” compares the extra cost only, against the 2 September value-date curve — not the rate itself, which fell with the market.

The rate fell a rupee. The cost of waiting went up at the long end.

Booked rates came down at every date, but almost entirely because spot fell 104.5 paise. Strip that out and the picture splits: one-month carry fell from 2.41–2.65% to 2.26–2.52%, while twelve-month carry rose from 2.89–2.91% to 3.07–3.09%. The curve steepened by roughly 14 paise at a year and 5 paise at six months.

That is a genuine change of shape, not noise, and it points the same way the flow argument does. A market that expects the rupee to stay strong does not charge more to defer. The forward market is pricing a wider rate gap over the coming year, not a narrower one.

What this means for each side

Importers. Two things went your way at once: spot fell more than a rupee and short-dated carry got cheaper. One-week cover at 94.4492 costs 4.2 paise and spans every catalyst on Friday — that is close to free insurance against the exact event we think is most likely to move against you. Beyond three months the argument reverses: cover at 95.05 already prices in our ceiling for this week, so you would be pre-paying for a level we rate under one in three.

Exporters. The whole curve came down with the market, so booked rates are lower everywhere. But the extra you earn by booking ahead is intact and improving: six-month cover at 95.8200 is 143 paise above today’s market, twelve-month at 97.2925 is 290 paise above, and both got better this week. For anything settling beyond three months, the forward gives you a materially better rate than the spot market plausibly will.

Where the forward curve crosses our markers

Read the last row carefully. An importer covering beyond three months is already paying above the level whose break we would treat as proof that the downtrend is over. Past that point, the argument this note is having about direction has already been settled by the forward market — and settled against the rupee.

The risk we rate highest, and why

The flow ending. It is the only risk on the page with a published date rather than a probability, and it lands on the same day as the largest scheduled data release of the week. Everything the flow has been masking — $97 oil, foreign selling, an equity market down 8.85% for the year — is still in place underneath it.

What sharpens it further is the market’s own condition. Daily movement has fallen to about 25 paise against a normal 34, and Monday’s session moved almost nothing. Calm this deep at the bottom of a fast move historically ends with a larger move rather than with more calm. The direction is not predictable from the calm itself. The size of what follows usually is.

What we still cannot see

The RBI’s reserves and intervention record. Friday’s weekly print is the first place the FCNR inflows should become visible in the official numbers, which makes it more useful this week than it has been all year. Until then we cannot separate the swap flow from ordinary market smoothing, and we are not going to pretend otherwise.

Day-by-day FCNR conversion data. We know the total raised and we know the deadline. We do not know how much was converted when, which is why section 01 labels the link between the scheme and the move as inference rather than measurement.

Options market pricing. No implied volatility or risk-reversal data. This would tell us whether the options market reads the current compression as directionless or skewed, which is exactly the question this week turns on.

Desk summary

This week, 8 to 11 September. Mildly bearish, with the risk turning from Friday. Closing band 94.30–94.80, intraday reach 94.10–95.05. The rupee-positive case is front-loaded; the dollar-positive case is back-loaded.

One month out. The question is whether the rupee holds its gains once the flow that produced them has gone. The US rate decision on 15–16 September is the likely decider. One-month cover already sits inside the exporter’s first tranche.

Beyond. The forward market charges 95.05 at three months and 97.29 at a year, and raised both this week. Any longer view must be measured against those numbers, not against today’s rate.

A · Technical evidence

Market structure

7 Sep: O 94.3900 · H 94.4250 · L 94.3800 · C 94.4225. Range 4.5 paise. Position in 50-day range 7.5%; in 52-week range 74.7%. 50-day high 96.6725, low 94.2400. Changes by window: 1d −0.066% · 5d −0.778% · 15d −1.051% · 30d −2.216% · 45d −0.827% · 60d −1.397% · 90d −0.124% · 120d +2.269% · 180d +4.832% · 360d +8.121%.

Trend

ADX14 40.96, labelled STRONG_TREND, 10-day slope +22.13 (was 19.91 on 20 Aug). DI+ 10.47 versus DI− 45.68 — a gap of 35.2, the widest in the series. Moving-average alignment BEAR; exponential stack not bullish. Price below SMA20 (95.2957, −0.916%), SMA50 (95.5081, −1.137%), SMA100 (95.2223, −0.840%), EMA8 (94.8348), EMA21 (95.1848), EMA55 (95.2580), EMA100 (94.7544). Above SMA200 (93.0477, +1.478%) — the only structural support for the multi-month uptrend. Supertrend DOWN at 95.1200 (X). Parabolic SAR DOWN at 95.5146. Ichimoku: price BELOW_CLOUD — conversion 95.0062, base 95.0175, cloud edges 95.7406 / 95.5475.

Momentum

RSI14 28.15 · RSI7 14.85 · OVERSOLD · no divergence. Only 1.24% of 889 sessions have printed RSI14 below 30, and the series low is 25.67. MACD −0.25221 below signal −0.12816; histogram −0.12406 and still falling. Stochastic K 10.33 / D 13.11. Williams %R −89.67 · CCI20 −176.52 · TSI −24.09. ROC12 −1.389% · ROC30 −2.216%.

Volatility

ATR14 0.2459 (24.6 paise) = 72.2% of its 50-day average. ATR20 0.2641. ATR as % of price 0.26. HV20 2.84% · HV60 4.17% · HV120 6.10% — HV20 is 47% of HV120. Parkinson 1.89% · Garman-Klass 1.73% · Rogers-Satchell 1.60%. Volatility percentile of history 44.0, state CONTRACTING. Bollinger bandwidth 0.01785, percentile 65.7, squeeze not active, position in band −0.013 (BELOW_LOWER). 1-day VaR 95: −0.548%. Expected shortfall: −0.869%.

Support and resistance zones

Clustering tolerance 0.1133 = 0.35 × average daily range. 16 zones identified in total.

Note the gap. Between 94.1350 and 93.1978 there is no identified zone at all — a 0.99-rupee vacuum. This is why the bearish scenario carries the smallest probability and the largest distance.

Fibonacci reliability

Of eight nearby levels, seven have been tested at least five times. The most reliable is the 60-day 61.8% retracement at 95.1043, which held on 70% of 10 tests. Nearest support, the 120-day 50% at 94.3400, held 60% of 5 tests; average move after a hold +0.52%, after a break −0.73%. Nearest resistance, the 252-day 23.6% at 94.5923, held only 33% of 6 tests — this level is more often crossed than respected, which is why 94.62 rather than 94.59 is our warning marker.

---

B · Historical and quantitative evidence

Statistical ranges, anchored on 94.4225

Published split by measurement basis. The trading window is four sessions, Tuesday to Friday.

Derived from 885 observations of this pair’s own history. The published closing band of 94.30–94.80 sits at the 50% band on the downside and between the 50% and 68% bands on the upside — deliberately asymmetric, reflecting the flow expiry. The published intraday reach of 94.10–95.05 uses the conditional upside excursion (95.0749) rather than the unconditional 75th (94.847), because comparable oversold situations have travelled further up than a general sample would suggest.

Conditional scenarios and probabilities

Engine output: bullish 30% / base 40% / bearish 30%. Regime STRONG_TREND, regime prior base 42%, historical tilt +0.32, structural tilt −0.70 (five structural supports for downside, none for upside). Published weights: 32 / 40 / 28. One adjustment made, and stated:

+2 to bullish, −2 from bearish. The engine’s positive historical tilt this week rests mainly on the after_momentum_below_35 stream (81.5% higher at five sessions, 27 observations), which is computed from closing prices and therefore survives the narrow-bar problem that contaminated last week’s compression reading. The range-based streams that are contaminated — unusually_narrow_range_day at 53% and unusually_small_daily_range at 47% — are near coin flips in either direction, so stripping them barely moves the weighted result. Separately, the FCNR expiry is a dated, one-directional catalyst that the historical engine has no way of seeing.

Weighted historical follow-through — seven usable streams, 470 observations

Weighted result: 50% at 1 day, 55% at 3 days, 58% at 5 days, 61% at 10 days. Lean classified NO_CLEAR_LEAN at 1 and 3 days, HIGHER at 5 and 10.

The market-condition analogue stream is excluded, not down-weighted. Searching 669 candidate days, best similarity was 53.9 against a 60 threshold and average similarity 30.8, classified VERY_LOW_SAMPLE. Today’s condition has no genuine historical precedent, so its 80% reading carries no information despite looking like evidence.

Desk back-tests on the current condition

Run directly on the 889-row series, since the engine does not publish these:

Two things follow. First, the direction of resolution has historically been upward, which supports the bounce case. Second, a full return to the middle band at 95.2957 within this week is a 20–25% proposition at best — which is why our ceiling is 95.05 and not 95.30. Note also that the mid-band is mechanically falling: if spot simply holds, it drops to roughly 95.05 by Friday as the 95.6–95.7 prints roll out of the twenty-day window.

Runs of consecutive closes below the lower band: 14 in the series, median 2 days, mean 1.86, maximum 4. Today is day 3. On duration alone the condition is close to expiring. The caveat that matters: none of the 22 precedents occurred with ADX above 35, and strong trends walk the lower band far longer than a general sample suggests.

Level-touch probabilities

From 20 comparable historical situations: nearest resistance zone centre 94.5089 touched 85% within five sessions and 90% within ten; nearest support zone centre 94.2400 touched 40% and 50%; 20-day low 94.2675 touched 40% and 50%; 20-day high 95.7675 touched 10% and 25%.

This asymmetry — 85% up versus 40% down — is the single cleanest number in the file, and it is the arithmetic basis for telling exporters to sell into strength rather than wait.

Treasury excursion metrics

---

C · Cross-currency and flow context

EUR/INR cross-rate integrity — last week’s failure is resolved

Last week we suspended EUR/INR publication after the 28 August quote failed the leg-consistency check by 55.7 paise. This week it passes. Quoted 109.6225 against USD/INR × EUR/USD of 109.6340 — a gap of −0.01%, against a 60-day mean of +0.008% and a standard deviation of 0.253%. Assessment: CONSISTENT. EUR/INR analysis can resume.

Correlation and decomposition

USD/INR to EUR/INR is 0.518 over 30 days, 0.549 over 60 and 0.593 over 120. USD/INR to EUR/USD is −0.056 over 30 days and effectively zero at 120. The dollar-index framing therefore remains structurally down-weighted for this pair, as corrected two cycles ago.

Decomposing EUR/INR’s 0.707% five-session fall: 93.6% came from the rupee leg and 6.4% from the euro leg. EUR/INR is currently a rupee story, not a euro story, and USD/INR is the pair to watch for both. Over twenty sessions the two legs moved in opposite directions and largely cancelled — the rupee strengthened 0.825% while the euro gained 0.472% against the dollar, netting EUR/INR down only 0.116%.

---

D · Sources, gaps and method

Sources

Desk quantitative and scenario engines v3.0 / v1.0, run 7 September at 09:06 IST. USD/INR series 889 rows, 2 January 2023 to 7 September 2026; all integrity checks passed, no duplicate dates, no invalid OHLC. Desk forward calculator, standard month-wise rates, spot value date 9 September 2026. FII/DII daily net figures for 31 August to 4 September. Real-time commodity futures and world market indices feeds. RBI announcements of 8 June and 14 August 2026 on the special swap facility, and contemporaneous reporting of the $52.3 billion figure. Prior week’s published outlook of 31 August for all week-on-week comparisons.

Data-quality note

Two faults in this week’s file, both handled per published method and neither affecting the levels in sections 01 to 06. First, 3 and 4 September carry identical OHLC and identical momentum readings to five decimals — a repeated print rather than two sessions. Second, the 7 September bar was captured at 09:06 and shows a 4.5-paise range against a normal 25, the third consecutive week an early capture has understated a session. The closing level is corroborated by the live forward-calculator quote of 94.3925/94.4025, so spot is anchored on the confirmed close; the bar is excluded from all volatility, candle-pattern and analogue computation. The engine run time moves to post-close from the next cycle.

Data marked unavailable — flagged, not estimated

Now resolved this cycle: forward curve (received in full) · FII/DII daily series · Brent and WTI absolute prices · Indian and global equity indices · gold, silver and volatility gauges · EUR/INR cross-rate integrity.

Still absent: RBI FX reserves and intervention record — Friday’s print is the first place the FCNR inflows should surface, making it more valuable this week than at any point this year. Day-by-day FCNR conversion data — we have the total and the deadline, not the timing. USD/INR implied volatility and risk reversals — would settle whether the options market prices this compression as directionless or skewed. India monthly trade balance — the oil-to-currency mechanism remains directionally certain and unquantified. Series completeness — 72 business days (7.49%) are missing across the full series.

Per client instruction, US Treasury yields are excluded from this cycle. No rate-differential attribution has been performed on the forward curve and none should be inferred from section 05.

Open signal conflicts

Engine evidence quality scores HIGH on data quality, indicator agreement and historical agreement, and LOW on historical sample; overall engine confidence MEDIUM. We publish directional confidence Medium and range confidence Medium.

Standing method notes

Levels are drawn from Fibonacci retracements, moving averages, volatility bands, channels, pivots, Ichimoku, swing points and round numbers, then clustered at 0.1133 tolerance. Historical analogues use a weighted robust-z nearest-neighbour search over backward-looking features only, across 669 candidate days; candidates must be at least 20 bars old so outcomes are complete, with a ±5-day exclusion so one episode cannot fill the list. The Ichimoku lagging line is excluded to avoid look-ahead bias. The engine is volume-free by design. Seasonality carries only 3–4 years per month, is rated VERY_LOW_SAMPLE, and has not been used. Any latest bar below the 5th percentile of range and captured within two hours of the open is excluded from volatility, candle-pattern and analogue computation. No rate, level, probability or event in this report has been invented.