Dalal Street Blogs provides information for educational purposes only and does not constitute investment advice. Please consult a SEBI-registered investment adviser before making any investment decisions.
CAUTION: Dalal Street Blogs publishes market news and educational content only. We do not promote, recommend or endorse any stock, and nothing here is a solicitation to buy or sell securities. Please consult a SEBI-registered adviser before investing.
Trading Strategy

Optimal f Position Sizing: The Growth-Maximising Bet Size and the Drawdowns It Brings

3897
By Dalal Street Blogs EditorialUpdated 10 min read

Most traders spend months on entries and about ten minutes on how much to put in each trade. Optimal f position sizing flips that around. It asks one blunt question of your trade history: what fixed fraction of the account, staked again and again, would have grown it the fastest? The answer is useful. The drawdowns that come with it are often brutal, and that second part is the bit most people skip.

The idea comes from Ralph Vince, and it’s a close cousin of the Kelly criterion. We’ll go through what optimal f actually measures, how to calculate it from a list of trades, how to turn it into a lot size, and why almost nobody should trade at the full number.

In this article
  1. Where Optimal f Comes From
  2. The Three Building Blocks: Biggest Loss, HPR and TWR
  3. A Worked Example: Finding Optimal f From Ten Trades
  4. Turning Optimal f Into a Position Size
  5. The Drawdowns That Come With Optimal f
  6. Why Optimal f Is Fragile in Practice
  7. How to Use Optimal f Sensibly
  8. A Quick Spreadsheet Method
  9. The Bottom Line

Where Optimal f Comes From

The Kelly criterion works out the bet size that maximises long-run growth when you know the probability of winning and the payoff on each side. For a simple bet that wins a fraction g of the stake with probability p and loses a fraction l with probability q, the Kelly fraction is p/l minus q/g.

That’s fine for a coin toss. Trading is messier. Wins come in all sizes, losses come in all sizes, and there’s no neat “probability of winning” that holds steady. Vince’s fix was to skip the formula and work straight from the actual sequence of trade results. Instead of plugging in averages, you test every possible fraction against the real trades and see which one compounds best.

The two methods agree when the game is simple. Take a coin toss where heads wins ₹2 and tails loses ₹1, each with probability 0.5. Kelly gives 0.5/1 minus 0.5/2, which is 0.25. In his March 2008 article on geometrical implications in money management, Vince uses the same game and shows the optimal f curve peaking at 0.25 too.

The Three Building Blocks: Biggest Loss, HPR and TWR

Optimal f needs three pieces. None of them is hard, but each one has a twist.

1. The biggest loss

Look through your trade list, measured per unit traded (one lot, or 100 shares, whatever you use consistently), and find the single worst loss. Everything else is scaled against it. This is the anchor of the whole method, and as we’ll see later, also its weakest point.

2. The holding period return (HPR)

Vince defines a quantity he calls f$, which is the biggest loss divided by f, with the sign flipped so it comes out positive. It’s the amount of account equity you set aside for each unit you trade. Then each trade’s HPR is 1 plus the trade’s profit or loss divided by f$.

In the coin toss at f = 0.25, f$ works out to ₹4. So you make one ₹1-risk bet for every ₹4 in the account. A win of ₹2 is a 50% gain on that ₹4, giving an HPR of 1.5. A loss of ₹1 is a 25% hit, giving an HPR of 0.75. Those are the exact numbers Vince uses.

Here’s the twist people miss: f is not the percentage of capital you put into a trade. It’s the fraction of your account that would vanish if your biggest historical loss showed up again. At f = 0.25, a repeat of the worst trade costs 25% of equity. That’s a very different thing from “I’m putting 25% of my money in.”

3. The terminal wealth relative (TWR)

Multiply all the HPRs together and you get the TWR, the multiple your account would have ended at after the whole sequence of trades. Take the Nth root of TWR, where N is the number of trades, and you get the geometric mean HPR: the average growth per trade, compounded.

Vince is pointed about this. The plain arithmetic average of returns tells you very little, he argues, because what you trade with today depends on what you made or lost before. Compounding is what matters. Optimal f is simply the value of f that gives the highest TWR, which is the same as the highest geometric mean.

Optimal f isn’t how much of your account goes into a trade. It’s how much of your account disappears if your worst trade turns up again.

A Worked Example: Finding Optimal f From Ten Trades

Let’s use a hypothetical swing trader, call her Meera, with ten closed trades. Each figure is the profit or loss for one unit of her position. These numbers are made up for illustration.

TradeP&L per unit (₹)HPR at f = 0.20
1+4,5001.300
2-3,0000.800
3+2,0001.133
4-1,5000.900
5+6,0001.400
6-2,5000.833
7+1,0001.067
8-3,0000.800
9+3,5001.233
10-1,0000.933

Five winners, five losers, and a net of +₹6,000 per unit across the ten trades. The biggest loss is ₹3,000, which happened twice.

To find optimal f, we step f from 0.01 to 0.99 in 0.01 increments, a plain brute-force search, and compute TWR each time. Here’s what a few of those values look like:

ff$ (equity per unit, ₹)TWRGeometric mean HPR
0.0560,0001.09021.0087
0.1030,0001.15821.0148
0.1520,0001.20061.0185
0.2015,0001.21571.0197
0.2512,0001.20261.0186
0.3010,0001.16231.0152
0.407,5001.00901.0009
0.506,0000.78570.9762
0.605,0000.53480.9393
0.803,7500.12600.8129

The peak sits at f = 0.20, where TWR is 1.2157. At 0.19 it’s 1.2149 and at 0.21 it’s 1.2153, so the top of the curve is fairly flat. Sizing at 0.20 would have turned ₹1 into about ₹1.22 over these ten trades.

Now look at the right side of the table. Same trades, same edge, same +₹6,000 per unit. At f = 0.40 the account barely moves. At 0.50 it loses about 21%. At 0.80 it’s down to roughly an eighth of where it started. Nothing about the strategy changed. Only the size did.

Vince makes the same point with his coin toss. Even in a game where you win ₹2 and lose ₹1 at even odds, his curve shows that staking more than 0.5 loses money, and keep doing it and you’ll go broke with near certainty. In his words, every market and every technique can be a loser if you bet the wrong quantity.

Turning Optimal f Into a Position Size

Once you have f, the sizing step is simple. Divide account equity by f$.

Say Meera’s account is ₹5,00,000. At f = 0.20, f$ is ₹15,000 (₹3,000 divided by 0.20). So she trades one unit for every ₹15,000 of equity: 5,00,000 divided by 15,000 is 33.3, which rounds down to 33 units.

Check what that means for the worst case. If a ₹3,000 loss per unit hits all 33 units, she loses ₹99,000, about 19.8% of the account. That’s f in action: roughly a fifth of equity gone on one bad trade. And she’s sizing like this deliberately, because it’s “optimal”.

Because equity changes after every trade, f$ stays fixed but the number of units changes. When the account grows, she trades more units. When it shrinks, fewer. That’s what makes it a fixed fractional method, the same family as the broader risk management rules most traders start with, just with a much more aggressive fraction.

The Drawdowns That Come With Optimal f

Here’s where the maths stops being pretty. Let’s run Meera’s ten trades in order on her ₹5,00,000 account at full optimal f, and again at half of it.

After tradeEquity at f = 0.20 (₹)Equity at f = 0.10 (₹)
16,50,0005,75,000
25,20,0005,17,500
35,89,3335,52,000
45,30,4005,24,400
57,42,5606,29,280
66,18,8005,76,840
76,60,0535,96,068
85,28,0435,36,461
96,51,2535,99,048
106,07,8365,79,080

At full optimal f, the account went from ₹7,42,560 after trade 5 down to ₹5,28,043 after trade 8. That’s a 28.9% drop from peak, inside a sample of just ten trades with a positive edge. Imagine sitting through that in real money. Most people would abandon the system right at the bottom.

At half f, the worst peak-to-trough fall was 14.8% (₹6,29,280 to ₹5,36,461). The final TWR was 1.1582 against 1.2157. So half f gave up a bit more than a quarter of the gain (15.8% against 21.6%) while roughly halving the deepest drawdown. That trade-off is the whole argument for fractional sizing.

Stack losses back to back and it gets worse. At f = 0.20, two maximum losses in a row leave 0.8 times 0.8, or 64% of the account. Three in a row leave 51.2%. At f = 0.10 the same three losses leave 72.9%. A risk of ruin calculation helps put probabilities on streaks like these.

The strategy didn’t change between f = 0.20 and f = 0.50. Only the size did, and that alone turned a 22% gain into a 21% loss.

The Kelly literature says much the same thing. Wikipedia’s summary notes that gamblers often bet half or quarter Kelly to cut the chance of ruin, reduce volatility and allow for errors in their estimates, that full Kelly brings high drawdowns, and that betting more than the Kelly amount increases the risk of ruin. Optimal f inherits every one of those problems.

Why Optimal f Is Fragile in Practice

It’s fitted to the past

Optimal f is calculated on trades you’ve already taken. The f that would have been best on the last 50 trades is not the f that will be best on the next 50. Since the curve falls off steeply to the right of the peak, overestimating f costs far more than underestimating it. In our table, going from 0.20 down to 0.10 cut TWR to 1.1582. Going up to 0.40 cut it to 1.0090.

The biggest loss isn’t the biggest loss

Everything scales off the worst trade in your sample. But the worst trade in your sample isn’t the worst trade possible. A gap-down on a results day, a circuit hit, or a stop that fills far below where you placed it can produce a loss bigger than anything in your history. If Meera sized at f = 0.20 against a ₹3,000 worst case and then took a ₹6,000 loss per unit, that single trade would cost 40% of the account, not 20%.

The edge itself can shrink

Kelly-style sizing assumes you know your edge. The Kelly criterion article points out that if you overestimate your true probability of winning, the bet size drifts away from optimal and risk of ruin goes up. A trading system’s edge moves around too. If it weakens, the “optimal” f you computed is now too big. Knowing your trading expectancy and checking it regularly matters more than any single sizing number.

Dispersion eats growth

Vince also shows that the geometric mean HPR is closely approximated by the arithmetic mean HPR and the variance of the HPRs. His conclusion: cutting the spread of your returns improves compound growth in the same way as raising the average return. For a trader, that’s a strong case for smaller size and steadier results over swinging for the peak.

How to Use Optimal f Sensibly

We don’t think optimal f is useless. It’s best read as a ceiling, not a target. A few practical ways to use it:

  • Treat it as the upper limit. If your calculation says 0.20, you know that sizing above it hurts both growth and survival. Anything sensible sits well below.
  • Trade a fraction of it. Half or a quarter of optimal f is a common middle ground, the same logic as half or quarter Kelly. In our example, half f kept most of the growth with about half the drawdown.
  • Pad the biggest loss. Use a worst-case loss larger than anything in your sample, say one that accounts for gaps beyond your stop. That automatically shrinks f$ sizing.
  • Use a decent sample. Ten trades is enough to show the mechanics, but far too few to trust the answer. More trades across different market conditions give a more honest curve.
  • Re-run it. Recalculate periodically as new trades come in, and size down if the result drops.
  • Stress-test the sequence. Shuffle your trade order many times to see how bad drawdowns could have been. A Monte Carlo simulation does exactly that.

A Quick Spreadsheet Method

You don’t need code. Put your trade P&L per unit in column A. Put the biggest loss, as a positive number, in a fixed cell. In column B, compute the HPR for a given f as 1 plus (trade divided by biggest loss) times f. Multiply all the HPRs with a PRODUCT function to get TWR. Then build a small table that repeats this for f from 0.01 to 0.99, and pick the f with the highest TWR. As our table shows, TWR climbs to a peak and then falls, so once it starts dropping as f rises, you’ve passed the top.

Then, before you trade a single rupee at that number, halve it and look at the drawdown again.

The Bottom Line

Optimal f answers a narrow question very well: given these exact trades, what fixed fraction would have compounded the fastest? It’s a useful lens because it shows how sharply results swing with size alone, and how quickly an over-sized account goes backwards even with a real edge. But the growth-maximising number comes with drawdowns most traders can’t stomach, and it rests on a sample and a worst-case loss that the market can and will exceed. Use it to find the ceiling, then size comfortably below it.

Disclaimer: This article is for educational purposes only and is not investment advice. The trader and all trade figures in the examples are hypothetical. Trading involves risk of loss; consult a SEBI-registered investment adviser before making investment decisions.

Leave a Reply

Join the conversation

Commenting is open to registered readers. Sign up with just your email — no password needed, and it takes under a minute.

Sign up to comment Log in