Chapter 8 · Planning a Financial Life

8.2 Costs, Taxes and Behaviour: What Really Drives Your Returns

The return on the factsheet is not the return you keep. Measure how costs compound, how the rate and timing of tax change growth, why harvesting a loss is mostly a deferral, how probability explains loss aversion, recency and overconfidence, and how a few lines of arithmetic expose a "guaranteed return" scheme.

📍 The situation

Farhan and Divya, both 30, work in the same office in Chennai. Each starts a ₹10,000 monthly SIP for 25 years in a fund holding the same broad, hypothetical index portfolio. Divya buys a low-cost plan and leaves it alone. Farhan buys a plan with a higher expense ratio through an app that also sends "hot tips"; he trades a few stocks on the side, moves money into whatever did best last year, and stops his SIP for a while after every fall. The same week, the office WhatsApp group forwards a message: "Alpha Wealth Club — assured 3% every month, capital fully safe, 18,000 happy members, refer a friend and earn 1% more!"

Twenty-five years later, how far apart will Farhan and Divya be — and how much of the gap comes from costs, from taxes and from behaviour? Why does paying tax later beat paying the same tax every year? Why does checking a portfolio every day feel so bad, and why does last year's best fund so often disappoint? And can arithmetic alone prove that Alpha Wealth Club cannot keep its promise? This last section of the course answers these questions with tools from every chapter: compounding (§1.1), XIRR (§1.3, Measuring Returns), fund costs (§2.2, Mutual Funds by the Numbers), lognormal returns (§3.3, Random Walks and Monte Carlo) and diversification (§4.1–§4.2).

⚠️ About the numbers on this page

Tax rates on this page are written as symbols ($t$, $\tau$) and, where a number is needed, as illustrative values — they are not the current rates. Tax rules for interest, capital gains, holding periods, exemptions, set-off of losses and repurchases change often; check the Income Tax Department and a qualified adviser for the current position. Expense ratios, brokerage, STT and exchange charges are also illustrative — check AMFI, SEBI and exchange circulars and your broker's tariff. All people, funds and schemes are hypothetical; nothing here is investment or tax advice.

Costs Compound Too

Every rupee of cost — a fund's expense ratio, brokerage, statutory charges, the bid–ask spread (§2.1), an advisory fee — is taken out of the portfolio and therefore also stops compounding. A cost that looks small per year becomes large over a lifetime, for exactly the same reason that returns do.

📘 Definition 8.7 — Gross return, cost drag and net return

The gross return $R$ is what the underlying portfolio earns in a year. The cost drag $c$ is the total yearly cost as a fraction of the assets: the expense ratio (§2.2), plus trading costs, plus any fees charged on assets. The net return is approximately $R-c$; for costs deducted continuously through the year (like an expense ratio) the exact growth factor is about $(1+R)e^{-c}$, and the two agree to within a tiny margin for the costs met in practice.

📐 Theorem 8.9 (Formula) — The compounding of costs

Over $T$ years at gross return $R\gt0$ and cost drag $c$ (with $0\lt c\lt R$):

  1. The fraction of final wealth lost to costs is $$1-\left(\frac{1+R-c}{1+R}\right)^T\approx1-e^{-cT/(1+R)}.$$
  2. The fraction of the gain lost, $\dfrac{(1+R)^T-(1+R-c)^T}{(1+R)^T-1}$, is larger than the fraction of wealth lost, and both grow towards 100% as $T\to\infty$.
  3. In any one year, costs take the fraction $c/R$ of the return — so the same cost takes a bigger share of a lower return.

Why. (1) The ratio of net to gross wealth is $\left(1-\frac{c}{1+R}\right)^T$, and $\log(1-u)\approx-u$ for small $u$. (2) Write $a=(1+R)^T$, $b=(1+R-c)^T$, with $a\gt b\ge1$. The two fractions are $\frac{a-b}{a-1}$ and $\frac{a-b}{a}$, and dividing by the smaller $a-1$ gives the larger number. As $T\to\infty$, $b/a\to0$. (3) is the definition. ∎ This is the general form of the expense-ratio drag of §2.2: there, two plans of the same fund differed only by their costs and ended in the ratio $e^{-\Delta c\,T}$, whatever the market did.

✏️ Example 1 — Three cost levels over 25 years
₹10 lakh is invested for 25 years in portfolios that each earn an assumed gross 11% a year, with total costs of 0.5%, 1.5% and 2.5% a year. Find the final values, the share of wealth lost and the share of the gain lost in each case.
Gross $10\times1.11^{25}\approx$ ₹135.85 lakh (gain ₹125.85 lakh).
Net $10\times1.105^{25}\approx$ ₹121.35 lakh; $10\times1.095^{25}\approx$ ₹96.68 lakh; $10\times1.085^{25}\approx$ ₹76.87 lakh.
Shares lost Wealth: 10.7%, 28.8% and 43.4%. Gain: 11.5%, 31.1% and 46.9%. The approximation $1-e^{-cT/1.11}$ gives 10.7%, 28.7% and 43.1%.
Read it A cost of 2.5% a year — less than a quarter of the 11% return each year — takes almost half the gain over 25 years. Costs are certain; returns are not. Lowering a cost is one of the few return improvements an investor fully controls.

Trading costs. A trader pays costs every time a position is opened and closed. If the whole capital is turned over $m$ times a year and each round trip costs a fraction $k$ (brokerage, statutory charges such as STT, exchange fees, stamp duty, the spread and slippage — all illustrative and changing), the capital is multiplied by $(1-k)^m$ a year on top of whatever the trades earn.

📐 Theorem 8.10 (Formula) — Turnover drag

With $m$ round trips a year of the whole capital at cost $k$ each, the yearly cost factor is $(1-k)^m$, so the trading cost drag is $$1-(1-k)^m\approx mk\quad(mk\text{ small}).$$ A strategy that earns gross $R$ before costs nets $(1+R)(1-k)^m-1$.

Why. Each round trip leaves the fraction $1-k$ of the capital; $m$ of them multiply. By the binomial theorem, $(1-k)^m\ge1-mk$, with near equality when $mk$ is small. ∎

✏️ Example 2 — Farhan's side trading
Farhan trades ₹5 lakh, turning over the whole amount about 4 times a month (48 round trips a year) at an illustrative all-in cost of 0.1% per round trip. (a) What is the yearly drag? (b) If his trades earned a gross 12% a year before costs, what would ₹5 lakh become in 10 years with and without the costs? (c) What gross return must he earn just to match an index fund netting 11%?
(a) $(0.999)^{48}\approx0.9531$: a drag of about 4.69% a year ($mk=4.8\%$).
(b) Without costs $5\times1.12^{10}\approx$ ₹15.53 lakh. With costs the net return is $1.12\times0.9531-1\approx6.75\%$, and $5\times1.0675^{10}\approx$ ₹9.61 lakh — about ₹5.9 lakh less.
(c) $(1+R)\times0.9531=1.11$ gives $R\approx16.5\%$ a year before costs — a very high bar for trades made on tips.
Check The costs are charged on every trade whether it wins or loses; §7.3 (Backtesting, Costs and Algorithmic Trading) shows how the same drag turns many "profitable" backtests into losers.

Taxes: Rate, Timing and Tax-Loss Harvesting

Tax takes a share of the return, but when it is paid matters as much as its rate. Some income — like interest on a fixed deposit — is generally taxed every year as it accrues. A capital gain is generally taxed only when the asset is sold ("realised"), so the tax money stays invested until then. We compare the two with symbolic rates.

📘 Definition 8.8 — Taxed every year vs taxed on realisation
  • Taxed every year at rate $t$: each year's return $R$ is reduced to $R(1-t)$ (think of interest added to income and taxed at the investor's slab rate).
  • Taxed on realisation at rate $\tau$: the investment compounds untaxed, and on sale the tax is $\tau\times(\text{sale value}-\text{cost})$. Holding-period rules, exemptions and indexation (where they apply) change $\tau$ and the taxable gain — check the current rules.
  • The after-tax return is the constant yearly rate that turns the amount invested into the after-tax amount; the real after-tax return also removes inflation by the Fisher relation of §1.1.
📐 Theorem 8.11 (Formula) — After-tax growth and the value of deferral

For ₹1 invested for $T$ years at a pre-tax return $R\gt0$:

  1. Taxed every year at $t$: $\ V_{\text{yr}}=\big(1+R(1-t)\big)^T$.
  2. Taxed on realisation at $\tau$: $\ V_{\text{end}}=1+\big((1+R)^T-1\big)(1-\tau)$.
  3. If $t=\tau$ (with $0\lt t\lt1$) and $T\ge2$, then $V_{\text{end}}\gt V_{\text{yr}}$: deferring the same tax always helps, and the advantage grows with $T$ and $R$.

The real after-tax return is $\dfrac{1+R_{\text{after tax}}}{1+\pi}-1$.

Proof of (3). Fix $R$ and $T$ and regard both values as functions of the tax rate $s\in[0,1]$. $V_{\text{end}}(s)=(1+R)^T-s\big((1+R)^T-1\big)$ is a straight line. $V_{\text{yr}}(s)=(1+R-Rs)^T$ is strictly convex in $s$ for $T\ge2$ (its second derivative is $T(T-1)R^2(1+R-Rs)^{T-2}\gt0$). Both equal $(1+R)^T$ at $s=0$ and both equal $1$ at $s=1$. A strictly convex function lies strictly below the straight line joining two of its points, so $V_{\text{yr}}(s)\lt V_{\text{end}}(s)$ for $0\lt s\lt1$. ∎ Intuitively, under deferral the tax you will eventually pay is itself invested and earning for you in the meantime — an interest-free loan from the government.

✏️ Example 3 — Same pre-tax return, three tax treatments
₹10 lakh earns a pre-tax 8% a year for 15 years. Find the after-tax value and after-tax return if (a) it is taxed every year at an illustrative $t=30\%$, (b) it is taxed only at the end at $\tau=30\%$, (c) it is taxed only at the end at an illustrative $\tau=12.5\%$. (d) For a fixed deposit at 7% taxed yearly at 30%, with 6% inflation, find the real after-tax return.
Pre-tax $10\times1.08^{15}\approx$ ₹31.72 lakh.
(a) $10\times(1+0.08\times0.7)^{15}=10\times1.056^{15}\approx$ ₹22.64 lakh: an after-tax return of 5.60%.
(b) $10\,[1+(1.08^{15}-1)\times0.7]\approx$ ₹25.21 lakh, i.e. $(2.5205)^{1/15}-1\approx6.36\%$ — the same 30% tax, paid later, adds ₹2.56 lakh.
(c) $10\,[1+(1.08^{15}-1)\times0.875]\approx$ ₹29.01 lakh: 7.36% a year.
(d) After tax $7\%\times0.7=4.9\%$; real $1.049/1.06-1\approx-1.04\%$. The deposit is "safe", but it loses purchasing power at these illustrative rates (§1.1).
📘 Definition 8.9 — Tax-loss harvesting

Tax-loss harvesting means selling an investment that stands at a loss in order to realise the loss, so that it can be set off against realised gains (and, where the rules allow, carried forward), and then — if the investor still wants that exposure — reinvesting in the same or a similar asset. Which losses can be set off against which gains, for how many years they can be carried forward, and whether an immediate repurchase is allowed or treated specially are matters of tax law that differ between countries and change; check the current Indian rules before acting.

Why it is mostly a deferral. Selling at the lower price and buying back resets the cost to that lower price. The tax saved today, $\tau L$ on a harvested loss $L$, is matched by a larger taxable gain — and an extra tax of about $\tau L$ — when the new holding is finally sold. The benefit is the time value of that deferred tax, plus a permanent gain only in special cases: when the later gain is taxed at a lower rate or falls under an exemption, or when the investment is never sold.

✏️ Example 4 — Harvesting a ₹1 lakh loss
Divya has realised capital gains of ₹1 lakh this year. Another fund she holds, bought for ₹5 lakh, is now worth ₹4 lakh. Take an illustrative capital-gains rate $\tau=12.5\%$, ignore exemptions, and assume the rules allow the set-off. She harvests the loss, reinvests the ₹4 lakh at once in a similar fund, and will sell in 10 years. (a) How much tax does she save now? (b) How much extra tax will she pay in 10 years? (c) What is the benefit worth today if money earns 10% a year?
(a) The ₹1 lakh loss cancels the ₹1 lakh gain: $0.125\times1{,}00{,}000=$ ₹12,500 saved now.
(b) Her cost is now ₹4 lakh instead of ₹5 lakh, so the eventual gain is ₹1 lakh larger: about ₹12,500 more tax in year 10 (at the same rate).
(c) She gains ₹12,500 now and loses ₹12,500 in 10 years: worth $12{,}500\,(1-1.1^{-10})\approx$ ₹7,681 today — real, but much less than the "₹12,500 saved" headline.
Costs and cautions Two transactions cost spreads, any exit load and perhaps a few days out of the market; a similar fund is not identical; and rules on repurchases may limit the strategy. Harvesting is worth doing carefully, but it is a timing tool, not free money.
Interactive 1 — Where Does the Return Go? Costs, Taxes and Inflation

Under an assumed constant gross return, the blue curve is the portfolio with no costs or taxes, purple is after costs, green is after costs and tax (if sold in that year, or taxed every year), and the red dashed curve is the green one in today's rupees. Grey is the money put in. Tax rates are symbolic inputs — try the same rate charged every year and at the end. Play runs the years.

12.0%1.5%
12.5%6.0%25
no costs, no tax—
after costs—
after costs and tax—
in today's rupees—
share of gross gain lost: costs / tax—
real after-everything return—

Behaviour: The Probability Behind the Biases

Most investing mistakes are not arithmetic errors; they are reasonable-feeling reactions to randomness. Each of the classic biases has a probability calculation behind it — and seeing the calculation is the best defence.

📘 Definition 8.10 — Five behavioural biases
  • Loss aversion: a loss is felt more strongly than an equal gain. A simple model values an outcome $X$ as $X$ if $X\ge0$ and $\lambda X$ if $X\lt0$, with $\lambda\gt1$ (experiments often find values around 2). Myopic loss aversion is loss aversion combined with frequent checking.
  • Recency bias: giving too much weight to the latest results — buying last year's top fund, or expecting a recent trend to continue.
  • Overconfidence: overrating one's own skill or information, often after a lucky streak; it leads to over-trading.
  • Base-rate neglect: judging a case on striking evidence ("beat the market three years running") while ignoring how common the underlying type (genuinely skilled funds) is.
  • Regret aversion: choosing so as to avoid the pain of having been wrong — e.g. refusing to sell a loser, or piling into the asset that just did best.
📐 Theorem 8.12 (Formula) — How often you see a loss

If a portfolio follows geometric Brownian motion with drift $\mu$ and volatility $\sigma$ (§3.3), and you look at it every $\Delta$ years, the probability that a given look shows a loss since the previous look is $$P(\text{loss})=\Phi\!\left(-\frac{\nu\sqrt\Delta}{\sigma}\right),\qquad\nu=\mu-\tfrac12\sigma^2\gt0.$$ It is close to $\tfrac12$ for small $\Delta$ and falls towards $0$ as $\Delta$ grows.

Why. Over $\Delta$ years the log return is normal with mean $\nu\Delta$ and standard deviation $\sigma\sqrt\Delta$ (§3.3). A loss means a negative log return: $P=\Phi\!\left(\frac{0-\nu\Delta}{\sigma\sqrt\Delta}\right)$. ∎ The drift grows like $\Delta$ but the noise only like $\sqrt\Delta$: over short periods noise dominates, over long periods drift does. A loss-averse investor who checks daily therefore experiences an almost even stream of gains and losses — and feels the losses double.

✏️ Example 5 — Checking frequency and the feeling of losing
Model an equity index fund with $\mu=12\%$, $\sigma=18\%$ (illustrative, as in §3.3), so $\nu=10.38\%$. (a) Find the probability that a look shows a loss if you look every day (Δ = 1/252), month, year, 5 years and 10 years. (b) With $\lambda=2$, the average "felt" value of one look is $E[X]-(\lambda-1)E[X^-]$, where $X$ is the return since the last look and $X^-$ its loss part. Using the lognormal formulas, this is about −0.38% (daily), −0.61% (monthly), +0.72% (quarterly) and +9.9% (yearly). Interpret.
Day $\nu\sqrt\Delta/\sigma=0.1038\times0.0630/0.18\approx0.0363$: $P\approx48.6\%$.
Longer Month: $z\approx0.167$, $P\approx43.4\%$. Year: $z\approx0.577$, $P\approx28.2\%$. Five years: $z\approx1.290$, $P\approx9.9\%$. Ten years: $z\approx1.824$, $P\approx3.4\%$.
(b) Every look has a positive expected return, yet with $\lambda=2$ an investor who looks monthly or more often feels a loss on average; the felt value turns positive only for gaps of about two months or more (about 0.17 years by solving numerically). Daily checking means about 122 felt losses a year out of 252 looks.
Interpret The investment is the same; only the frequency of looking differs. Frequent checkers are more likely to sell after falls — locking in the losses that a longer horizon would usually have absorbed. Looking less often is not ignorance; it is choosing the $\Delta$ at which the signal beats the noise.
Interactive 2 — How Often Do You Look? Seeing Losses in a Rising Market

Ten years of a simulated index fund (seeded GBM, daily steps). Each dot is one look at the portfolio: green if it is up since the previous look, red if it is down. Change how often you look and how strongly you feel losses (λ). The readouts compare this path with the formula of Theorem 8.12 and with 200 simulated paths. Play draws the decade day by day.

2.012.0%18%
looks so far—
looks showing a loss (this path)—
formula Φ(−ν√Δ/σ)—
200 simulated paths—
average felt value per look—
value of ₹100 after 10 years—

Recency, luck and base rates. Every year some funds top the tables. Before reading skill into a streak, ask how many streaks pure luck would produce, and how common skill is to begin with.

📐 Theorem 8.13 (Rule) — Lucky streaks and base rates
  1. If $N$ funds have no skill, so that each lands in the top quartile of a year's ranking with probability $\tfrac14$ independently each year, then the expected number in the top quartile $k$ years running is $N/4^k$, and the probability that at least one does it is $1-(1-4^{-k})^N$.
  2. (Bayes.) If a fraction $b$ of funds is skilled, a skilled fund beats its benchmark in a year with probability $p_s$ and an unskilled one with probability $p_u\lt p_s$, independently across years, then $$P(\text{skilled}\mid k\text{ wins in a row})=\frac{b\,p_s^k}{b\,p_s^k+(1-b)\,p_u^k}.$$

Why. (1) Each fund's streak has probability $4^{-k}$; expected counts add (linearity of expectation), and the chance that no fund has a streak is $(1-4^{-k})^N$ by independence. (2) is Bayes' theorem with prior $b$ and likelihoods $p_s^k$ and $p_u^k$. ∎ The base rate $b$ matters as much as the evidence: if skill is rare, most streaks belong to the lucky.

✏️ Example 6 — Last year's star and Farhan's winning streak
(a) Of 400 equity funds with no skill at all, how many would you expect to be top-quartile three years running, and how likely is it that at least one is? (b) Suppose 10% of funds are skilled; a skilled fund beats the benchmark in a year with probability 0.65, an unskilled one (after its costs) with probability 0.45. What fraction of funds that beat the benchmark three years running are skilled? (c) Farhan wins 13 of his 20 trades and decides he has an edge. If his true win rate were 50%, how likely is 13 or more wins?
(a) $400/64=6.25$ funds expected; $P(\text{at least one})=1-(63/64)^{400}\approx99.8\%$. A "consistent top performer" list is guaranteed even in a world without skill.
(b) $p_s^3=0.2746$, $p_u^3=0.0911$: $$P=\frac{0.1\times0.2746}{0.1\times0.2746+0.9\times0.0911}\approx25\%.$$ Three in four "three-year winners" are lucky unskilled funds, because unskilled funds are nine times as common.
(c) $P(X\ge13)$ for $X\sim\text{Binomial}(20,0.5)$ is about 13.2% — roughly one trader in eight with no edge does this well. In 100 coin-flip trades the longest winning streak averages about 6 (by simulation, about 55% of such records contain a streak of 6 or more), which feels like skill but is not evidence of it.
Moral Short records cannot separate skill from luck (§7.3 makes the same point with t-tests on strategies). Recency bias buys the lucky; overconfidence keeps trading after a lucky run — and Example 2 shows what the trading costs.
📐 Theorem 8.14 — A diversified portfolio is never the best and never the worst

If a portfolio holds assets $1,\dots,n$ with fixed weights $w_j\ge0$, $\sum w_j=1$, then in every period $$\min_jR_j\ \le\ \sum_jw_jR_j\ \le\ \max_jR_j,$$ with strict inequalities whenever the returns are not all equal and at least two weights are positive.

Why. A weighted average with non-negative weights lies between the smallest and largest values averaged; if two assets with positive weight have different returns, it differs from both extremes. ∎ This is diversification seen through regret: a 60/30/10 equity/debt/gold mix in a year when they return +25%, +7% and −5% earns $15+2.1-0.5=16.6\%$ — 8.4 points behind the best asset, so there is always something to regret, but never the full regret of having put everything into the worst one. Regret aversion pushes people to chase whichever asset just won; a fixed allocation, rebalanced (§4.2), is a pre-commitment against that.

📈 Market case — From 12% on the factsheet to what Farhan and Divya keep

Both invest ₹10,000 at the start of every month for 25 years (₹30 lakh in all). The portfolio's gross return is assumed constant at 12% a year ($i=(1+R)^{1/12}-1$). Illustrative assumptions: Divya's total costs 0.5% a year, Farhan's 1.5%; tax $\tau=12.5\%$ on the whole gain at the end (exemptions ignored); inflation 6%. For Farhan we also assume a behaviour gap of 1.5 points a year: his mistimed switching and pauses (below) cost him that much relative to the fund.

stepreturn usedcorpus before taxafter taxafter-tax XIRRin today's ₹real XIRR
gross portfolio12%₹170.2 lakh₹152.7 lakh11.31%₹35.6 lakh5.01%
Divya: costs 0.5%11.5%₹157.3 lakh₹141.4 lakh10.82%₹32.9 lakh4.55%
costs 1.5%10.5%₹134.4 lakh₹121.3 lakh9.84%₹28.3 lakh3.63%
Farhan: + behaviour gap9.0%₹106.5 lakh₹97.0 lakh8.38%₹22.6 lakh2.25%
  1. Costs. The extra 1% a year costs ₹20.0 lakh after tax (₹141.4 vs ₹121.3 lakh); the full 1.5% costs about 21% of the gross corpus and about 26% of the gross gain — Theorem 8.9, for a SIP.
  2. Tax. At the end, $\tau$ is charged on (corpus − ₹30 lakh): between ₹9.6 and ₹17.5 lakh here. Deferral helps: had the same 12.5% been charged on each year's gain instead, every row would be lower (Theorem 8.11).
  3. Inflation. Dividing by $1.06^{25}\approx4.29$ turns ₹1.5 crore into about ₹35.6 lakh of today's purchasing power. Every rupee figure in a 25-year plan should be read this way.
  4. Behaviour. Where does a "behaviour gap" come from? Take a fund that returns +30%, −25%, +20% in three years: its CAGR is $(1.3\times0.75\times1.2)^{1/3}-1\approx5.37\%$. An investor who puts in ₹1 lakh at the start of each year earns an XIRR of about 4.37%. One who puts in ₹1 lakh, adds ₹2 lakh after the good year, and sells everything after the bad year (for ₹2.475 lakh) has an XIRR of about −13.6% in the same fund. The fund's return measures the fund; your XIRR (§1.3) measures you.
  5. Total. From a 12% factsheet assumption Divya keeps a real 4.55% a year and Farhan 2.25%. In today's rupees her corpus is about ₹10.3 lakh larger — almost entirely from costs and behaviour, the two items they controlled.

The mathematics, not a recommendation. All four rows use the same market. Costs, tax timing and behaviour are the parts of return an investor can influence; the market is not. The table's exact numbers depend on the assumptions — the method is what carries over.

Red Flags: Testing a "Guaranteed Return" Offer

Every chapter of this course has said the same thing in a different language: higher expected return comes with higher risk (§4.2), returns vary (§3.1), and costs and taxes are certain. An offer that promises a high, smooth, guaranteed return contradicts all three at once. A few lines of arithmetic make that contradiction concrete.

📘 Definition 8.11 — Ponzi scheme and pyramid scheme

A Ponzi scheme pays its promised "returns" to existing investors out of money from new investors, not from genuine earnings. A pyramid scheme pays members mainly for recruiting new members. Both can pay reliably for years, and both must collapse when new money stops growing fast enough. Such schemes are illegal; in India, depending on the activity, the securities regulator (SEBI), the RBI and other authorities deal with unregistered investment and deposit schemes, and regulators publish ways to check registrations and report suspicious offers.

📐 Theorem 8.15 — The arithmetic of a Ponzi scheme

A scheme with no real earnings promises a return $r$ per month on all money deposited, paid at the end of each month, and keeps the principal.

  1. Constant inflow $N$ a month: the month's payout exceeds the month's new money once $t\gt1/r$, and the cash pile $C_t=Nt-rN\frac{t(t+1)}{2}$ becomes negative once $t\gt\frac2r-1$.
  2. Growing inflow $N(1+g)^{s-1}$ in month $s$: the payout in month $t$ exceeds the next month's inflow once $(1+g)^t\gt\dfrac{r}{r-g}$ when $g\lt r$. Payouts can keep up for ever only if new money grows at least at the promised rate, $g\ge r$, for ever.

Proof. (1) After $t$ months the deposits are $D_t=Nt$ and the month-$t$ payout is $rNt$, which exceeds $N$ when $t\gt1/r$. Total payouts are $rN(1+2+\cdots+t)=rN\frac{t(t+1)}{2}$, so $C_t=Nt\big(1-\frac{r(t+1)}{2}\big)\lt0$ when $t+1\gt2/r$. (2) Deposits are a geometric sum, $D_t=N\frac{(1+g)^t-1}{g}$. The payout $rD_t$ exceeds $N(1+g)^t$ when $(1+g)^t\left(\frac rg-1\right)\gt\frac rg$, which is the stated condition for $g\lt r$. As $t\to\infty$, $rD_t/[N(1+g)^t]\to r/g$, which is below 1 only if $g\gt r$. ∎ A promise of 3% a month needs new money to grow at least 42.6% a year, compounding for ever: in 30 years, about 42,000 times the starting inflow.

✏️ Example 7 — Can Alpha Wealth Club pay 3% a month?
(a) What yearly return is "3% a month"? Compare with an illustrative 7% on a government bond. (b) If the club has no real earnings and attracts a constant ₹1 crore of new money a month, when do monthly payouts exceed new money, and when does the cash run out? (c) If instead every member must bring in 6 new members to earn the referral bonus, how many "levels" of recruiting exceed 100 crore people?
(a) $1.03^{12}-1\approx42.6\%$ a year, doubling every $\log2/\log1.03\approx23.4$ months — six times the illustrative bond yield, with "no risk". If the promoter really had a riskless 42.6%, he could borrow from banks at a fraction of that and would not need strangers' money. That alone is a red flag.
(b) $1/r\approx33.3$: from month 34 the month's payout exceeds the ₹1 crore of new money. $2/r-1\approx65.7$: the cash pile is gone in month 66 — after five and a half years of perfect payments. A long record of payments is not evidence that a scheme is genuine.
(c) Level $k$ needs $6^k$ members: $6^k\gt10^9$ when $k\gt\log10^9/\log6\approx11.6$, so by the 12th level the scheme would need more recruits than there are people in the country. Late joiners — the large majority — must lose.
Lesson The question is never "has it paid so far?" but "where can this return come from?" If the answer is "from the next investor", the arithmetic of Theorem 8.15 decides the ending.
📐 Theorem 8.16 (Rule) — The smoothness test

If a genuine strategy's monthly returns are independent and each month is positive with probability $p\lt1$, the probability of $n$ positive months in a row is $p^n$. Under GBM with drift $\mu$ and volatility $\sigma$, $$p=\Phi\!\left(\frac{\nu/12}{\sigma/\sqrt{12}}\right),\qquad\nu=\mu-\tfrac12\sigma^2 .$$

Why. Independence multiplies the probabilities; the formula for $p$ is Theorem 8.12 with $\Delta=1/12$, read from the other side. ∎ A track record with no losing months, or with the same return every month, is not impressive — it is improbable.

✏️ Example 8 — A track record that is too good
A "portfolio management" pitch shows 36 consecutive months of exactly +2% and says the strategy invests in equities and derivatives. (a) For a genuine equity strategy with $\mu=12\%$, $\sigma=15\%$, find the probability of a positive month and of 36 positive months in a row. (b) Even if a strategy were positive in 80% of months, how likely is a 36-month unbroken run? (c) What other numbers in the pitch should worry you?
(a) $\nu=0.12-0.01125=0.10875$; $p=\Phi\!\left(\frac{0.10875/12}{0.15/\sqrt{12}}\right)=\Phi(0.209)\approx0.583$. Then $p^{36}\approx3.6\times10^{-9}$ — about 4 in a billion.
(b) $0.8^{36}\approx0.0003$, about 1 in 3,000.
(c) +2% every month is 26.8% a year with essentially zero volatility: an infinite Sharpe ratio (§3.2), something no genuine equity or derivatives strategy produces. Identical returns usually mean the numbers are being set, not earned.
Lesson Genuine risky returns are bumpy. Smoothness plus high return is the signature of fabricated statements or of a hidden risk that has not yet shown itself (§6.2: an option seller's steady income can end in one large loss).
🧭 The red-flag checklist for any "guaranteed return" offer
  1. Compare the return with the risk-free rate. A "guaranteed" return far above what a government bond pays has no explanation except hidden risk or fraud (Example 7a).
  2. Ask where the return comes from — which assets, earning what, held by whom. "A secret strategy", "forex arbitrage", "crypto mining", "algo trading" with no verifiable detail is not an answer.
  3. Look at the shape of the record. No losing months, identical monthly returns or round-number returns fail the smoothness test (Theorem 8.16).
  4. Check registration with the relevant regulator (SEBI for investment advisers, portfolio managers, brokers and mutual funds; the RBI for banks and NBFCs; IRDAI for insurers) on the regulator's own website, not on a link the promoter sends.
  5. Follow the money. Payments to a personal account or wallet, cash, or "pay a fee to withdraw your profits" are red flags; so is any request for OTPs, passwords or remote access to your phone.
  6. Referral rewards for bringing in new investors point to a pyramid (Example 7c).
  7. Pressure and secrecy — "limited slots", "today only", "don't tell your bank" — are there to stop you doing this checklist.
  8. Social proof is cheap. Screenshots of profits, celebrity photos and "18,000 happy members" can be faked, and early members of a Ponzi really are paid (Example 7b).
  9. Test withdrawal terms. Lock-ins that keep extending, delays and new conditions on withdrawal are how collapses begin.
  10. If you are unsure, do not invest — and report the offer to the authorities. The cost of missing a genuine opportunity is small; the cost of joining a fraud can be everything you put in.
⚠️ Common mistakes and myths
  1. "1% a year is too small to matter." Over 25 years 1.5% took 29% of the wealth and 31% of the gain at 11% (Example 1).
  2. "Cost is a share of the return." It is a share of the assets, taken every year whether the return is high, low or negative.
  3. "Frequent trading is free if brokerage is zero." Statutory charges, spreads and slippage remain, and they multiply with turnover (Theorem 8.10).
  4. "Only the tax rate matters." The same 30% charged every year cost ₹2.56 lakh more than at the end over 15 years (Example 3).
  5. "Tax-loss harvesting saves the whole tax." It mostly defers it: ₹12,500 "saved" was worth about ₹7,681 (Example 4).
  6. "My fund returned 15%, so I earned 15%." Your XIRR depends on when your money went in and out; chasing returns can turn a positive fund return into a personal loss (Market case).
  7. "Checking daily keeps me safe." It shows a loss almost half the time (Example 5) and invites selling at the wrong moment.
  8. "Three years of top performance proves skill." Luck alone produces such streaks among hundreds of funds (Example 6).
  9. "It has paid every month for years, so it must be real." A Ponzi scheme at 3% a month can pay perfectly for five years (Example 7).
  10. "Guaranteed and high." In a market where risk-free money earns a few percent, those two words together are the clearest red flag.

Exercises

All rates, costs and tax rates are illustrative. Use the formulas of this section and of §1.1–§1.3; round sensibly.

I
Costs.
  • 1 ₹5 lakh is invested for 20 years at a gross 10% a year. Find the final value with no costs and with 1% a year of costs, and the share of wealth lost.
  • 2 In part 1, what share of the gain is lost?
  • 3 Estimate the share of wealth lost to a 0.75% cost over 30 years at 10% with the exponential approximation, and compare with the exact value.
  • 4 A trader makes 36 round trips a year of his whole capital at an all-in cost of 0.15% each. Find the yearly cost drag.
  • 5 If the trades of part 4 earn a gross 15% a year before costs, what is the net return?
  • 6 A trader's costs take 5% of his capital each year. What gross return must he earn to match an index fund that nets 11%?
  • 7 What fraction of the yearly return does a 1% cost take when the gross return is 6%? When it is 12%? What does this say about costs on low-return products?
1 $5\times1.1^{20}\approx$ ₹33.64 lakh; $5\times1.09^{20}\approx$ ₹28.02 lakh; share lost $1-(1.09/1.1)^{20}\approx16.7\%$.
2 Gains ₹28.64 lakh and ₹23.02 lakh: $\dfrac{1.1^{20}-1.09^{20}}{1.1^{20}-1}\approx19.6\%$ of the gain.
3 $1-e^{-0.0075\times30/1.1}\approx18.50\%$; exact $1-(1-0.0075/1.1)^{30}\approx18.56\%$.
4 $(0.9985)^{36}\approx0.9474$: a drag of about 5.26% a year ($mk=5.4\%$).
5 $1.15\times0.9474-1\approx8.95\%$.
6 $(1+R)\times0.95=1.11$: $R\approx16.8\%$.
7 $1/6\approx16.7\%$ and $1/12\approx8.3\%$ of the return. On low-return products (debt funds, liquid funds) the same percentage cost takes a much larger share of what the investor earns.
II
Taxes and harvesting.
  • 1 ₹10 lakh earns 7% a year for 10 years, taxed every year at $t=30\%$. Find the final value and the after-tax return.
  • 2 The same pre-tax 7%, but taxed only at the end at $\tau=30\%$. Find the final value and the after-tax return.
  • 3 Repeat part 2 with $\tau=20\%$.
  • 4 With 5% inflation, find the real after-tax return in part 1.
  • 5 An investor harvests a ₹40,000 loss against gains taxed at $\tau=20\%$ and expects to sell the repurchased holding in 8 years. Find the tax saved now and the present value of the benefit at 9%.
  • 6 Explain why harvesting is mostly a deferral, and give one situation in which it creates a permanent saving.
1 $10\times1.049^{10}\approx$ ₹16.13 lakh; after-tax return 4.9%.
2 $10\,[1+(1.07^{10}-1)\times0.7]\approx$ ₹16.77 lakh; $(1.6770)^{1/10}-1\approx5.31\%$.
3 $10\,[1+(1.07^{10}-1)\times0.8]\approx$ ₹17.74 lakh; about 5.90% a year.
4 $1.049/1.05-1\approx-0.10\%$: after tax, the deposit barely keeps up with inflation.
5 Saved now: $0.2\times40{,}000=$ ₹8,000; an extra ₹8,000 is due in 8 years. Benefit today: $8{,}000\,(1-1.09^{-8})\approx$ ₹3,985.
6 Buying back resets the cost to the lower price, so the later taxable gain rises by the harvested loss and the tax saved now is paid later. It becomes permanent if the later gain is taxed at a lower rate or falls within an exemption, or if the holding is never sold (the deferral never ends).
III
Behaviour and probability.
  • 1 A fund follows GBM with $\mu=10\%$, $\sigma=20\%$. Find the probability that a look shows a loss over 1 month, 1 year and 10 years.
  • 2 With $\lambda=2$: over a year a fund is up with probability 0.7 by 18% on average and down with probability 0.3 by 10% on average. Find the average felt value. Over a month it is up with probability 0.55 by 3.5% on average and down with probability 0.45 by 3% on average. Find the felt value and compare.
  • 3 Of 500 funds with no skill, how many do you expect in the top quartile 3 years running? How likely is at least one?
  • 4 20% of funds are skilled; skilled funds beat the benchmark with probability 0.6 a year, unskilled with 0.4. Find $P(\text{skilled})$ after 3 straight wins and after 5.
  • 5 A trader wins 14 of 20 trades. If the true win rate is 50%, how likely is 14 or more wins?
  • 6 A 50/40/10 equity/debt/gold portfolio in a year with returns −15%, +8%, +20%: find its return, and how far it is from the best and the worst asset.
  • 7 A fund returns +20%, −10%, +15% in three years. An investor puts in ₹1 lakh at the start, adds ₹2 lakh after year 1, and sells everything at the end of year 2. Find the fund's 3-year CAGR and the investor's XIRR (annual cash flows).
1 $\nu=0.10-0.02=0.08$. Month: $\Phi(-0.08\times0.2887/0.2)=\Phi(-0.115)\approx45.4\%$; year: $\Phi(-0.4)\approx34.5\%$; 10 years: $\Phi(-1.265)\approx10.3\%$.
2 Year: $0.7\times18-2\times0.3\times10=12.6-6=+6.6\%$. Month: $0.55\times3.5-2\times0.45\times3=1.925-2.7=-0.775\%$. The expected monthly return is positive ($1.925-1.35=0.575\%$), but it feels negative; the yearly view feels positive.
3 $500/64\approx7.8$ funds; $1-(63/64)^{500}\approx99.96\%$.
4 3 wins: $\dfrac{0.2\times0.216}{0.2\times0.216+0.8\times0.064}\approx45.8\%$. 5 wins: $\dfrac{0.2\times0.07776}{0.2\times0.07776+0.8\times0.01024}\approx65.5\%$. Even five straight wins leave a one-in-three chance of luck.
5 $P(X\ge14)$, $X\sim\text{Bin}(20,0.5)$: about 5.8% — suggestive, but with many traders trying, someone always gets this.
6 $0.5(-15)+0.4(8)+0.1(20)=-2.3\%$: 22.3 points behind gold, 12.7 points ahead of equity (Theorem 8.14).
7 CAGR $=(1.2\times0.9\times1.15)^{1/3}-1\approx7.49\%$. The investor holds $(1.2+2)\times0.9=2.88$ lakh at the end of year 2; solving $-1-\frac{2}{1+x}+\frac{2.88}{(1+x)^2}=0$ gives $x\approx-3.0\%$. Same fund, opposite experience: a behaviour gap of about 10 points.
IV
Red flags.
  • 1 Convert "2% a month guaranteed" to a yearly rate and find its doubling time in months.
  • 2 A Ponzi scheme promises 2.5% a month with constant new money each month. When do payouts first exceed new money? When does the cash run out?
  • 3 New money grows 2% a month and the promised return is 4% a month. After how many months do payouts exceed new money?
  • 4 Each member must recruit 5 new members. How many levels until more than 100 crore people are needed?
  • 5 A genuine strategy is positive in 60% of months. How likely is a record of 36 positive months in a row?
  • 6 An offer says: "Fixed 18% a year, capital protected, only 20 slots left, pay to this UPI ID." List four checks from the checklist and what each would reveal.
1 $1.02^{12}-1\approx26.8\%$ a year; doubling time $\log2/\log1.02\approx35.0$ months.
2 $1/r=40$: payouts exceed new money from month 41. $2/r-1=79$: the cash pile is exactly zero after month 79 and negative in month 80.
3 $(1.02)^t\gt\dfrac{0.04}{0.04-0.02}=2$: $t\gt\log2/\log1.02\approx35.0$, so from about month 36.
4 $5^k\gt10^9$: $k\gt\log10^9/\log5\approx12.9$, so 13 levels.
5 $0.6^{36}\approx1.0\times10^{-8}$ — about 1 in 100 million.
6 (i) Compare 18% "fixed and protected" with the risk-free rate — far above it, so the risk is hidden. (ii) Ask what earns 18% and who holds the assets. (iii) Check registration on the regulator's own website. (iv) Payment to a personal UPI ID and "only 20 slots" are red flags in themselves — follow the money, and treat pressure as a reason to stop.
🎯 Practice corner NISM-styleCase studyInterview

E1. NISM-style The total expense ratio of a mutual fund scheme is (A) collected once a year from the investor's bank account (B) accrued daily and reflected in the NAV (C) charged only on redemption (D) charged only when the scheme makes a profit.

Reason Expenses are accrued every day as a fraction of net assets and are already deducted in the published NAV (§2.2) — which is why the investor never "sees" them, and why they compound (Theorem 8.9).
Answer (B).

E2. NISM-style Two investments earn the same pre-tax return and face the same tax rate. One is taxed every year; the other only when it is sold after 10 years. The second will normally (A) end with more money (B) end with less money (C) end with the same amount (D) end with more only if the market rises.

Reason Deferral leaves the tax money invested until the sale (Theorem 8.11(3)); with a positive return and $T\ge2$ the deferred value is strictly larger.
Answer (A).

E3. NISM-style Which feature of an investment offer is the strongest warning sign of a fraudulent scheme? (A) The NAV is published daily (B) The offer document describes the risks (C) "Assured 5% a month, plus a bonus for every friend you bring in" (D) Returns vary from month to month.

Reason 5% a month is about 79.6% a year, "assured", with recruitment rewards — a Ponzi/pyramid pattern (Theorem 8.15). (A), (B) and (D) are features of genuine, regulated products.
Answer (C).

E4. Case study ₹20 lakh is invested for 20 years in two plans of a portfolio with an assumed gross return of 10%: plan X has total costs of 1.2% a year, plan Y 0.3%. Tax at an illustrative $\tau=12.5\%$ is paid on the gain at the end; inflation is 5%. Find each plan's after-tax value, its value in today's rupees, and the difference.

Before tax X: $20\times1.088^{20}\approx$ ₹108.05 lakh; Y: $20\times1.097^{20}\approx$ ₹127.40 lakh.
After tax X: $108.05-0.125(108.05-20)\approx$ ₹97.04 lakh; Y: $127.40-0.125(127.40-20)\approx$ ₹113.97 lakh.
Today's ₹ Divide by $1.05^{20}\approx2.6533$: X ≈ ₹36.57 lakh, Y ≈ ₹42.96 lakh.
Difference ₹16.93 lakh after tax (₹6.38 lakh in today's rupees), from a cost gap of 0.9% a year. Tax narrows the rupee gap a little (it takes 12.5% of each plan's gain), but cannot close it.

E5. Case study A scheme with no real earnings promises 5% a month, paid at each month's end on all deposits. New money starts at ₹1 crore in month 1 and grows 3% a month. (a) From which month does the payout exceed the next month's new money? (b) Simulating the cash pile month by month, in which month does it run out? (c) What would new money need to do for the scheme to survive indefinitely?

(a) $(1.03)^t\gt\dfrac{0.05}{0.05-0.03}=2.5$: $t\gt\log2.5/\log1.03\approx31.0$, so the payout of month 31 already exceeds month 32's new money.
(b) Each month: cash += new money; deposits += new money; cash −= 5% × deposits. The cash first goes negative in month 52, when deposits total about ₹121.7 crore — over four years of flawless payouts.
(c) Grow at least 5% a month (about 79.6% a year) for ever (Theorem 8.15(2)). In 10 years that is a factor of $1.05^{120}\approx349$ — impossible for any real pool of savers.

E6. Interview "A fund manager has beaten the market ten years in a row. Surely that proves skill?" There are 2,000 managers, and assume none has any skill, so each beats the market in a year with probability ½ independently. How many ten-year streaks do you expect, and how likely is at least one? What would you ask next?

Count Expected streaks $2000/2^{10}\approx1.95$; $P(\text{at least one})=1-(1-2^{-10})^{2000}\approx85.8\%$.
Interpretation A ten-year streak is what luck usually produces somewhere in a large population. The question "did someone do it?" says nothing; "would this particular manager have done it?" needs the base rate of skill (Theorem 8.13) and evidence chosen before looking at the winners.
Next questions By how much did the manager win (a t-statistic on the excess returns, §7.3)? Was the benchmark appropriate? How many funds did the same firm close or merge (survivorship)? Does the edge have an economic explanation? Interview answer: "Not by itself — with 2,000 coin-flippers, I'd expect about two such records."
🧑‍🏫 For the teacher

Suggested plan — 2 lectures × 60 min.

  1. Costs and taxes. Open with the 📍 situation and ask the class to guess how much smaller Farhan's corpus will be. Theorem 8.9 with Example 1 (ask for a guess before showing 47% of the gain lost), turnover drag with Example 2, then Theorem 8.11 with its convexity proof drawn as a picture (a convex curve under a chord) and Example 3. Harvesting with Example 4. Project Interactive 1: set the same tax rate in both modes and switch.
  2. Behaviour and fraud. Theorem 8.12 and Example 5 with Interactive 2 — let a student choose the checking frequency and count red dots. Example 6 (luck and base rates) as a class vote first. Theorem 8.14 and the 📈 Market case table, built row by row. Finish with Theorem 8.15, Example 7 and the 🧭 red-flag checklist applied to a real (anonymised) forwarded message.

Class activity. A coin-flip tournament: every student flips a coin five times ("beat the market or not"); those with five heads stand up and are interviewed about their "strategy". Then compute the expected number of five-head records for the class size — and compare with how many stood up.

Misconceptions to watch for. Costs as a share of return; tax rate without timing; harvesting as a permanent saving; fund return = investor return; streaks as proof of skill; long payment records as proof of legitimacy.

Suggested homework. I(1, 2, 4, 6), II(1–5), III(1, 3, 4, 7), IV(1–5); Practice corner E4, E5, E6.

🛠️ Do it yourselfSpreadsheet mini-project · about 60–90 minutes

Project. Measure your own (or a hypothetical) "return leak": build a sheet that takes a SIP from a factsheet return to a real, after-cost, after-tax XIRR, and then measures a behaviour gap from actual cash flows.

Why it matters. Factsheets report the fund's return; your statement shows only rupees. The difference between the two — costs, taxes, timing — is invisible until you compute it, and it is the part you can change.

What to do.

  1. Inputs: monthly SIP B1, years B2, gross return B3, total cost B4, tax rate on gains B5 (symbolic — type the current rule yourself), inflation B6.
  2. Four corpora with =FV((1+r)^(1/12)-1,12*B2,-B1,0,1) for r = B3, B3-B4, and B3-B4-gap; after-tax value =V-B5*(V-B1*12*B2); today's rupees =V/(1+B6)^B2.
  3. XIRR: make a dated column of −B1 for each month and the after-tax value on the last date; =XIRR(values,dates). Real XIRR =(1+x)/(1+B6)-1. Check the first row against the 📈 Market case.
  4. Behaviour gap: take a fund's month-end NAVs (from the AMC or AMFI website) and two investors — a steady SIP, and a "chaser" who doubles the SIP after any 12-month return above 20% and stops it for a year after any 12-month return below −10%. Compute both XIRRs and the fund's CAGR over the same dates.
  5. Tax timing: add a column that taxes each year's gain at the same rate (V_next=V*(1+r*(1-t))) and compare with the deferred version — Theorem 8.11 on your own numbers.

Python version: from scipy.optimize import brentq; xirr=brentq(lambda r: sum(cf/(1+r)**(d/365) for cf,d in flows),-0.9,1), with flows a list of (amount, days from the first date).

Hand in: a one-page table like the Market case for your inputs, the chaser's and the steady investor's XIRR next to the fund's CAGR, and three sentences on which leak was largest.

Going further. Add a red-flag score sheet: for any offer you receive, record the promised return, the illustrative risk-free rate, the claimed monthly record (count losing months), registration status and payment channel — and compute the Ponzi collapse month of Theorem 8.15 for its promised rate.

Instructor note: with ₹10,000 a month, 25 years, 12% gross, costs 0.5% and 1.5%, tax 12.5% at the end and 6% inflation, the sheet should reproduce after-tax corpora of ₹141.4 and ₹121.3 lakh and XIRRs of 10.82% and 9.84%. Spreadsheet FV with type 1 means start-of-month instalments; forgetting it understates every corpus by the factor 1 + i.

🧪 Self-check quiz

0/5 correct

Q1. At a gross return of 10% a year, a cost of 1% a year over 30 years removes roughly what share of the final wealth?

Q2. With the same rate, why does tax paid only on sale beat tax paid every year?

Q3. For a fund with $\mu=12\%$, $\sigma=18\%$, the probability that a single day's look shows a loss is about

Q4. 10% of funds are skilled. A skilled fund beats its benchmark with probability 0.65 a year, an unskilled one with 0.45. Among funds that beat it three years running, the share that is skilled is about

Q5. A scheme with no real earnings promises 4% a month and receives the same amount of new money every month. Its cash runs out after about