Triangle Strategy: Near-Zero Correlation Gave Me 50% More Return with 12% Less Risk

A week ago I published my barbell thesis: two Indian stocks, two independent risk factors, a +0.25 correlation. HDFC Bank for compressed value, Muthoot Finance for mispriced growth. The framework held up. The numbers were clean. I was happy with it.

Then I ran one more correlation check. And the barbell stopped being good enough.

The problem wasn’t that the barbell was wrong. It was that I couldn’t find negative correlation anywhere in Indian equities. I screened 112 NSE stocks — 56 large-caps, 56 mid-caps — across 247 trading days. Zero negative correlations. Not one. The lowest positive pair I found was HDFC + Muthoot at +0.25, which is good by Indian standards but still means both stocks tend to move in the same direction about 60% of the time.

The +0.25 is real diversification within India. But it has a ceiling. Common macro factors — FII flows, rupee moves, oil prices, RBI policy — pull everything on the NSE in the same direction. You can reduce correlation within Indian equities. You can’t eliminate it.

So I asked a different question: what happens if I add a stock from a different country?

---

The Triangle: 3 Stocks, 2 Countries, Zero Overlap

I kept HDFC Bank and Muthoot Finance. They’re both still cheap, the theses are intact, and their +0.25 correlation is the best domestic pair I found. But I added a third vertex: NVIDIA.

Not because I wanted tech exposure. Because the correlation math demanded it.

I pulled 238 common trading days of daily return data across all three stocks — HDFC Bank (NSE), Muthoot Finance (NSE), and NVIDIA (NASDAQ) — covering March 2025 through March 2026. Here’s what came back:

The Correlation Matrix

	HDFC Bank	Muthoot Finance	NVIDIA
HDFC Bank	1.00	+0.23	-0.02
Muthoot Finance	+0.23	1.00	-0.02
NVIDIA	-0.02	-0.02	1.00

NVIDIA has essentially zero correlation with both Indian stocks. Not low positive. Not slightly negative. Zero. Different country, different currency, different sector, different macro drivers. The cross-border diversification delivered what 112 NSE stocks couldn’t.

And the number is stable. Here’s how the correlations held across quarters:

Quarter	HDFC↔Muthoot	HDFC↔NVIDIA	Muthoot↔NVIDIA
Q2 2025	+0.28	-0.16	-0.36
Q3 2025	+0.22	+0.07	+0.12
Q4 2025	+0.22	+0.08	+0.18
Q1 2026 (crisis)	+0.18	+0.06	+0.31

HDFC-Muthoot stays between +0.18 and +0.28 across all regimes — remarkably consistent. HDFC-NVIDIA stays near zero even during the Iran crisis. The only quarter where Muthoot-NVIDIA correlation spiked (+0.31) was Q1 2026, when everything sold off together during the worst of the crisis. Even then, the direction agreement was limited.

On 70% of trading days, at least one vertex moves in the opposite direction from the other two. All three stocks move the same way only 30.4% of the time. That’s what a triangle gives you that a barbell can’t.

Stress Test: When One Vertex Breaks

Numbers are one thing. How the portfolio actually behaves during pain is another. I isolated the worst single-day drops for each stock and measured what the other two did on those same days:

When this crashes...	HDFC moves	Muthoot moves	NVIDIA moves
NVIDIA > -3% (21 days)	-0.10% (flat)	+0.03% (flat)	—
Muthoot > -3% (12 days)	-0.73%	—	+1.41% (rises)
HDFC > -2% (5 days)	—	-0.99%	-0.01% (flat)

The Muthoot-NVIDIA line is the standout. When Muthoot drops more than 3%, NVIDIA rises an average of +1.41%. That’s not noise — it’s 12 independent observations. Different currencies, different sentiment drivers. Gold fear crushes Muthoot but doesn’t touch NVIDIA. AI capex selloffs hit NVIDIA but don’t affect Indian gold lending.

When one vertex breaks, the others hold. Sometimes they gain. That’s the entire point of the triangle.

Portfolio Math: Why 3 > 2

I backtested two portfolio constructions over the same 12-month period:

Portfolio	12M Return	Volatility	Return/Risk
60/40 HDFC+Muthoot (barbell)	+14.8%	18.9%	0.78
50/30/20 HDFC+Muth+NVDA (triangle)	+22.1%	16.7%	1.32
100% NVIDIA (for reference)	+50.8%	41.7%	1.22

The triangle delivered +50% more return with 12% less volatility. The return-to-risk ratio nearly doubled from 0.78 to 1.32. And notice: even 100% NVIDIA, despite returning +50.8%, has a worse risk-adjusted ratio (1.22) than the triangle (1.32). Diversification across near-zero-correlated assets is doing real mathematical work here.

The triangle isn’t three random stocks. It’s three stocks specifically chosen because their risk factors don’t overlap:

Vertex	Stock	Thesis	Risk Factor
Compressed Value	HDFC Bank	Largest private bank at 52W lows, accelerating earnings	Indian banking / macro
Mispriced Growth	Muthoot Finance	70% PAT growth, priced as gold proxy. 60% of earnings gold-independent	Gold / NBFC
AI Monopolist	NVIDIA	86% AI chip market, +73% revenue growth, forward PE 15.8x	US tech / AI capex

Three independent bets. Each can succeed even if the other two fail. That’s the upgrade from the barbell.

---

Vertex 1 — HDFC Bank: Compressed Value

I covered this in detail in the barbell post. The thesis hasn’t changed — if anything, it’s gotten stronger. Quick recap:

HDFC Bank is India’s largest private bank, trading near its 52-week low at a 30% discount to its own 10-year average P/E. The TTM P/E is 16.8x versus a historical average of 22-24x. Earnings are accelerating: -0.6% → +9.3% → +12.8% over the last three quarters. The HDFC Ltd merger integration is completing. The market is pricing in a story the numbers aren’t telling.

Quarter	PAT (₹ Cr)	YoY Growth	Signal
Mar 2025	19,285	+7.1%	Post-merger stabilization
Jun 2025	17,090	-0.6%	Seasonal trough
Sep 2025	20,364	+9.3%	Acceleration begins
Dec 2025	20,691	+12.8%	Inflection point

The trade is mean reversion. If earnings growth sustains at 12-15% and P/E reverts to even 20x (still below historical average), the math gives you 35-40% upside over 18-24 months.

The 5 HDFC Watchpoints

But I’m not blind to the risks. HDFC has returned +6.2% over 3 years while ICICI returned +53.2%. The stock first crossed ₹700 in January 2021 and is back at ₹780 today — five years of dead money. The merger compressed NIM from 4.1% to 3.35% and suppressed ROE from 18%+ to 14.6%.

Here are the five signals I’m watching to distinguish “compressed value” from “value trap”:

#	Watchpoint	Bull Trigger	Bear Trigger
1	NIM Recovery	NIM ≥ 3.50%	NIM stays ≤ 3.35%
2	ROE Trajectory	ROE ≥ 15.5%, trending up	ROE stays ≤ 14.5%
3	Bank Nifty Rebalancing	Passive selling overhang lifts after Mar 31	Price doesn’t recover post-rebalancing
4	Governance Resolution	No further board exits, RBI affirms	More resignations, RBI action, forensic audit
5	FII Selling Reversal	FII 3-day avg turns positive	FII ownership drops below 45%

The review date is April 2026, after Q4 FY26 results. If 4-5 of 5 bull triggers fire, the thesis is confirmed. If 0-1 fire, it’s a value trap and the capital gets redeployed.

---

Vertex 2 — Muthoot Finance: Mispriced Compounder

Muthoot’s PEG ratio is 0.22. That’s a P/E of 15.3x divided by 70% earnings growth. PAT doubled year-on-year in the December quarter. Revenue grew 47%. ROE is above 20%.

The market treats Muthoot as a gold proxy. It’s not. Here’s the earnings breakdown:

Earnings Driver	Contribution	Gold-Dependent?
Gold price appreciation (AUM uplift)	~40%	Yes
Loan book expansion	~30%	No — secular growth
Operating leverage from scale	~20%	No — structural
Branch network growth	~10%	No — secular growth

60% of Muthoot’s earnings growth is gold-independent. Even if gold corrects 15-20% and total growth halves to 35%, the PEG would be 0.44 — still very cheap. An ROE of 20%+ in a 15x P/E NBFC is a valuation anomaly the market hasn’t corrected because it can’t see past the gold label.

The Gold-Lag Entry Framework

Here’s something I found in the data that I haven’t seen written about anywhere: when gold crashes more than 2% in a single day, Muthoot’s pain doesn’t show up immediately. It lags.

On the day gold crashes (Day 0), Muthoot barely moves: -0.4% average. The real selling hits on Day 1 (-2.4% average) as analysts publish notes, LTV ratio concerns surface, and retail panic kicks in. The trough doesn’t arrive until Day 4-5.

Gold Drop (single day)	Severity	Expected Muthoot Impact (3-day)	Expected Trough (by Day 4-5)
2-3%	Mild	-2.3%	-2.9%
3-5%	Moderate	-4.4%	-4.5%
>5%	Severe	-7.2%	-8%+

The framework is simple: when gold crashes, don’t buy Muthoot on Day 0. Don’t buy on Day 1. Wait for the Day 3-5 trough window. Set limit orders using severity multipliers applied to Day 0’s closing price. Cancel if unfilled by Day 5 — the selling wave is over.

The Key Asymmetry

This is the really interesting part. The gold-Muthoot relationship isn’t symmetric:

Metric	Value
Upside beta (gold rises → Muthoot)	-0.02 (essentially zero)
Downside beta (gold falls → Muthoot)	+0.52 (follows gold down)
Gold drops >3%: Muthoot also drops	83% of the time
Gold rises >3%: Muthoot also rises	Only 62% of the time
Lagged correlation (gold today → Muthoot tomorrow)	+0.28 (strongest lag)

The market punishes Muthoot when gold drops but doesn’t reward it when gold rises. That’s the mispricing. 60% of earnings are gold-independent, but the stock behaves as if 100% are gold-dependent on the downside. Gold sentiment dips create buying opportunities for a business that’s growing from lending spreads, not just gold prices.

The 5 Muthoot Watchpoints

#	Watchpoint	Bull Trigger	Bear Trigger
1	PAT Growth Sustainability	Quarterly PAT growth stays >40% YoY	PAT growth decelerates to <25% for 2 consecutive quarters
2	Loan Book Growth (ex-Gold)	Gold loan book + personal/vehicle loan growth >20% YoY	Loan book growth stalls <10% while gold flatlines
3	NPA / Asset Quality	GNPA stays <2.5%, no regulatory red flags on LTV	GNPA spikes >4% or RBI tightens gold LTV rules
4	Gold Sensitivity Repricing	Stock stops dropping 2x gold on bad days	Stock continues -2x gold beta on downdswings, no re-rating
5	Competitive Position vs Manappuram	Muthoot maintains PE premium, AUM lead widens	Manappuram closes the gap on AUM while trading cheaper

---

Vertex 3 — NVIDIA: The AI Monopolist

NVIDIA doesn’t need a long thesis. Everyone knows the story: 86% of the AI chip market, +73% revenue growth, data center revenue larger than most tech companies’ entire business. The question isn’t whether NVIDIA is a great company. It’s whether you’re paying a fair price.

At $172.70, NVIDIA is 18.6% below its 52-week high. The forward P/E is 15.8x. That’s cheaper than Microsoft (28x), Apple (26x), Amazon (25x), and Alphabet (18x). The company growing fastest is priced cheapest. That disconnect is what makes this interesting.

Why NVIDIA Is Cheap Right Now

Three forces are pushing NVIDIA down simultaneously, and none of them are fundamental:

1. GCC sovereign wealth fund selling. Saudi PIF cut US equities from $56.6B (2021) to $12.9B (Q4 2025) — a 77% reduction that started before the Iran war. The war accelerated it. Physical destruction of Gulf infrastructure (1,800+ missiles hit UAE, Dubai real estate index down 30% in two weeks) created urgent capital needs for reconstruction. SWFs are pulling from US tech to fund domestic rebuilding. This is a forced liquidation, not a bearish thesis.

2. Geopolitical risk premium. US-Iran tensions create uncertainty. Markets hate uncertainty. Tech stocks with high multiples get hit first in risk-off moves, regardless of fundamentals.

3. AI narrative fatigue. Every cycle has this phase: from euphoria (“AI changes everything”) through skepticism (“Is anyone making money from AI?”) back to fundamentals (“Look at the actual revenue numbers”). We’re in the skepticism phase. Meanwhile, NVIDIA’s Q4 FY26 revenue was $68.1 billion. Up 73% year-over-year. Hyperscaler capex is hitting $600 billion in 2026. Jensen Huang projects a $1 trillion AI infrastructure market by 2027.

The Picks-and-Shovels Argument

Everyone debates which AI company will win downstream — OpenAI, Anthropic, Google, xAI. It doesn’t matter for the NVIDIA thesis. They all buy NVIDIA GPUs. The picks-and-shovels seller doesn’t need to know which miner finds gold. They just need miners to keep digging.

CUDA lock-in makes this structural, not cyclical. Switching from NVIDIA to AMD or custom silicon means rewriting millions of lines of code. The $600 billion in planned hyperscaler capex for 2026 is already committed. This isn’t speculative demand — it’s purchase orders.

The 5 NVIDIA Watchpoints

#	Watchpoint	Bull Trigger	Bear Trigger
1	Revenue Growth	Quarterly revenue growth stays >40% YoY	Revenue growth decelerates to <25% for 2 consecutive quarters
2	Hyperscaler Capex	Big 4 capex stays >$500B/yr combined	Capex cuts >15% from any 2 of the Big 4
3	CUDA Moat	No major customer announces full migration away from CUDA	Google TPU or AMD MI400 captures >25% training share
4	China/Export Risk	Export controls stabilize, no new restrictions	Full China GPU ban or broad allied export restrictions
5	Valuation Floor	Forward PE stays <25x with earnings revisions up	Forward PE expands >30x as estimates get cut

---

The Entry Framework: Simple Grid Sizing

The entry system is the same for all three stocks. Two variables: how far has it fallen from its 52-week high, and has it stabilized?

“Stabilized” means 2 consecutive sessions where the intraday low is higher than the previous session’s low. It’s a crude but effective filter against catching falling knives.

Fall from 52W High	Still Falling	Stabilized (2-day higher lows)
10-15%	Small tranche	Moderate tranche
15-20%	Moderate	Large tranche
20-25%	Large	Full tranche
25%+ or at 52W low	Full tranche	Maximum tranche

Exception: Stock at 52-week low during an exogenous shock = deploy immediately, no stabilization required. Quality stock at 52W low during geopolitical crisis is structural opportunity.

Hysteresis: Don’t Jump on the First Green Signal

Stabilization requires 2 consecutive sessions of confirmation. Don’t deploy large tranches the first day a stock stops falling. Require the low to hold for two sessions. This is the hysteresis rule — it prevents getting baited by dead-cat bounces.

The only override: stock hitting a 52-week low. That’s a structural signal strong enough to skip the wait.

The 5/5 Crisis Regime

I run a 5-signal regime detector that categorizes the macro environment:

Signal	Bearish Threshold	Current Reading
India VIX	> 22	22.81 — Bearish
Brent Crude	> $100	$106.77 — Bearish
INR/USD	> ₹95	₹93.65 — Bearish
FII 3-day avg	< ₹-2,000 Cr	₹-5,264 Cr — Bearish
Bank Nifty RSI	< 25	22 — Bearish

All 5 signals are bearish simultaneously for the first time. This is a 5/5 CRISIS regime. It escalated from RISK_OFF (2/5) to CRISIS (4/5) to full 5/5 in the span of four trading days.

In CRISIS regime, tranches auto-reduce. The grid says to deploy large, but the regime says the macro is hostile. The system resolves this by scaling down tranche sizes — still buying, but in smaller bites. The thesis drives the direction (buy into panic), the regime controls the sizing (don’t overcommit when everything is red).

---

Gold: The Structural Bull Case

Gold drives 40% of Muthoot’s earnings growth. If gold’s rally is temporary, Muthoot’s PEG of 0.22 is a trap. If it’s structural, it’s the cheapest way to ride a multi-year trend.

Gold went from $2,000 to $5,595 in 24 months. Five structural pillars support the rally, and none of them are about Iran:

1. Central bank buying. After the West froze $300 billion of Russian reserves in 2022, non-aligned central banks drew a simple conclusion: dollar assets can be weaponized. Over 1,000 tonnes purchased per year for three consecutive years. China, India, Turkey, Poland all stacking.

2. De-dollarization. USD share of global reserves fell from 70% to 59%. Gold fills the gap as a neutral reserve asset no government can freeze.

3. Dollar weakness. DXY fell more than 10% in 2025, hitting a 4-year low. Gold and the dollar are inversely correlated.

4. The 60/40 portfolio is broken. Stocks and bonds fell together from 2022-2025. Gold stepped in as the real portfolio hedge. Institutional allocations to gold are rising structurally.

5. Geopolitical risk is the new normal. Iran is the latest entry on a list that includes Ukraine-Russia, Taiwan tensions, and trade wars. The risk premium isn’t tied to any single conflict.

Roughly 70% of gold’s rally is structural. About 30% is Iran event-driven. Even if Iran de-escalates tomorrow, gold loses the 30% event premium and stabilizes around $4,000+. That’s still a massively supportive environment for Muthoot.

Iran de-escalation is bullish for gold-linked stocks. Oil drops (helping Indian macro and HDFC), gold stabilizes at structural levels (supporting Muthoot’s AUM), and risk-off unwinds (helping NVIDIA). All three vertices benefit. The triangle was built for this.

---

Scenario Matrix: The Triangle in 4 Futures

Scenario	Prob	Macro	HDFC	Muthoot	NVIDIA	Triangle
Bull	25%	Crisis resolves, rate cuts, FII return	+35%	+15%	+40%	+30%
Base	40%	Gradual normalization, gold firm, AI capex holds	+20%	+35%	+25%	+25%
Bear	25%	Oil crisis extends, gold surges, tech sells off	-10%	+25%	-15%	+2%
Worst	10%	Gold corrects AND banks weak AND AI capex cut	-20%	-15%	-25%	-20%

Three of four scenarios are profitable. In the Bear scenario, where oil crisis extends and tech sells off, Muthoot’s gold tailwind keeps the portfolio above water. In the Bull scenario, HDFC re-rates sharply and NVIDIA bounces on risk-on sentiment. In the Base case, all three contribute modestly. They take turns carrying the portfolio.

The Worst case requires gold correcting materially while Indian banking stays weak AND AI capex gets cut simultaneously. Those conditions are partially contradictory: if the US achieves enough geopolitical dominance to crash gold, India’s macro typically improves and HDFC rallies. If AI capex gets cut, the reason is usually recession — which sends gold higher. Breaking all three vertices at once requires a very specific and unlikely macro combination.

---

The Guiding Principle

I’ll buy what FIIs are dumping, but only at my price.

FII selling creates the opportunity. They’re selling Indian equities because of macro pressure — oil, rupee, geopolitics. They’re selling US tech because of risk-off sentiment and sovereign wealth fund liquidation. None of that changes the underlying businesses.

HDFC Bank’s earnings are still accelerating. Muthoot’s PAT still doubled. NVIDIA’s revenue still grew 73%. The businesses haven’t changed. The prices have. That’s the gap the triangle exploits.

The framework doesn’t require timing the bottom. It requires deploying capital at pre-defined prices, in pre-defined sizes, with stabilization confirmation. It’s systematic, not discretionary.

The Bottom Line

The barbell was two independent bets with +0.25 correlation. The triangle is three independent bets across two countries with near-zero cross-correlation on two of three pairs.

HDFC Bank: India’s best bank at a 30% P/E discount with earnings accelerating from -0.6% to +12.8% in three quarters. Mean reversion target: 35-40% over 18-24 months.

Muthoot Finance: PEG 0.22. PAT doubled. 60% of earnings are gold-independent. The market punishes it 2x on gold downswings but doesn’t reward it on upswings. Gold-lag rules let you buy the sentiment dips at statistically optimal prices.

NVIDIA: 86% AI chip market at a forward PE of 15.8x — cheaper than every big tech peer. The selling is forced (SWF liquidation) and sentiment-driven (AI fatigue), not fundamental. CUDA lock-in and $600B committed hyperscaler capex provide the floor.

The triangle: +22.1% backtested return, 16.7% volatility, 1.32 return/risk ratio. Three of four scenarios profitable. When one vertex breaks, the others hold or gain. That’s not diversification by accident — it’s diversification by design.

I screened 112 NSE stocks across 247 trading days. Zero negative correlations. The lowest positive was +0.25. Then I added one stock from a different country and got -0.02. The math is simple: the hedge Indian equities couldn’t provide, geography did.

---

Disclaimer: This is not investment advice and should not be construed as a recommendation to buy, sell, or hold any security. This post is a personal analytical framework based on publicly available data. The author is not a SEBI-registered investment advisor. The author may hold positions in securities discussed. All data points are from publicly available sources as of March 2026. Past performance does not guarantee future results. Please consult a SEBI-registered advisor before making any investment decisions.