Skip to content
Menu
Finabase Learn

Build and test a strategy Lesson 12 of 14

Strategies · Building and testing a strategy

Position sizing inside a strategy: how much, not what

See how the amount committed to each signal changes a strategy: one set of QQQ signals at nine sizes, the point where more size earns less, and volatility-scaled sizing.

On this page
  1. Size is a separate decision
  2. One signal list, nine sizes
  3. How much is too much: the Kelly idea
  4. Sizing by risk per trade
  5. Sizing by volatility
  6. Where sizing rules fail
  7. The methods compared
  8. How Finabase does it
  9. Frequently asked questions
  10. Summary
  11. Check your understanding
  12. Exercise

A strategy answers two questions. The first is when to be in the market and when to be out. The second is how much to commit each time. Most of the attention goes to the first question, and most of what a person actually experiences, the size of the gains and above all the size of the falls, is set by the second.

This lesson holds the signals fixed and changes only the size. It takes one list of buy and sell dates on QQQ and runs it at nine sizes, from a quarter of the account to four times the account with borrowed money. Then it looks at sizing that changes over time, by scaling the position to recent volatility. The sizing of a single trade from a stop distance is covered in Position sizing: how many shares to buy for a set risk; this page is about the strategy as a whole.

Size is a separate decision

Take two people who follow exactly the same rule on the same fund. One puts a quarter of the account into each signal. The other puts in all of it. They buy and sell on the same days, are right and wrong on the same trades, and have the same win rate. Their accounts look nothing alike.

That is why a backtest's return line means little until the size is stated. The standard test on the strategy pages of this library puts the whole account into the fund on a buy signal and all of it into cash on a sell. That is one choice among many, and usually not the one a careful person would make with real money.

Size can be set in three broad ways:

  • A fixed share of the account, such as half of it on every signal.
  • A fixed risk per trade: choose the loss that would be accepted if the stop is hit, such as 1% of the account, and let the stop distance set the number of shares.
  • A target level of volatility: hold more when the market is calm and less when it is violent, so the account's own swings stay roughly constant.

One signal list, nine sizes

The signals are those of the 10-month moving average rule on QQQ, explained in Moving average trend following: the 10-month rule, tested and read line by line in How to read a backtest report, line by line: hold the fund while its month-end close is above its 10-month average, otherwise hold cash.

  • Data. QQQ month-end closes, January 2000 to October 2026, 322 months.
  • Sizing. At each month-end the position is reset to a fixed multiple of the account. A multiple of 0.5 means half in the fund and half in cash. A multiple of 2 means the fund position is twice the account, with the second half borrowed.
  • Left out. Trading costs, interest on borrowed money, interest earned on cash, dividends and taxes. Leaving out borrowing costs makes every multiple above 1 look better than it could have been.

The table also runs the same sizes with no rule at all, simply holding the fund, to show what size does to a rougher ride.

Size (multiple of the account) Rule: return per year Rule: deepest fall Hold: return per year Hold: deepest fall
0.25 2.1% −10.3% 2.4% −30.6%
0.5 4.2% −19.8% 4.6% −53.4%
0.75 6.2% −28.5% 6.4% −69.8%
1 8.0% −36.5% 7.9% −81.1%
1.5 11.4% −50.6% 9.8% −93.5%
2 14.4% −62.4% 9.9% −98.1%
2.5 16.7% −71.9% 8.0% −99.6%
3 18.5% −79.7% 3.6% −99.9%
4 19.9% −90.4% wiped out −100%
QQQ monthly, 2000–2026: yearly return at each position size (no costs, no interest) 0% 5% 10% 15% 20% 0.25× 1× 1.5× 2× 2.5× 3× Buy and hold 10-month rule Share of capital in the fund when invested (1 = all of it, 2 = borrowed as much again) Return per year, %
QQQ monthly, 2000 to 2026: return per year at each position size, with no costs and no interest on borrowed money. The chart stops at three times the account; at four times, buying and holding was wiped out.

Below full size: a straight trade

Up to a multiple of 1 the picture is simple. Halving the size roughly halves both the return and the deepest fall. At half size the rule earned 4.2% a year with a deepest fall of 19.8%, against 8.0% and 36.5% at full size. Nothing is gained or lost in efficiency; the dial just turns the whole experience up or down. For someone who could not sit through a fall of 36.5%, a smaller size is a more direct answer than a different strategy.

Above full size: the return bends

Above 1 the two columns part. The fall keeps growing in step with the size. The return does not.

For buying and holding, twice the size earned 9.9% a year, only two points more than full size, in exchange for a fall of 98.1%. At 2.5 times the return was back down to 8.0%, at three times 3.6%, and at four times the account went to zero. QQQ's worst month in the sample was a loss of 26.2%. Four times that is more than everything.

Why more size can earn less

Compound growth is hurt by swings, and the damage grows with the square of their size. The arithmetic is easy to check:

Doubling the position doubles every monthly gain and every monthly loss. The average monthly return doubles too. But a month of −26.2% at twice the size is −52.4%, which needs a gain of 110% to repair, not twice the 35.5% that repairs the original loss. Each large loss costs more than a matching gain gives back, and the gap widens as size goes up. At some size the extra average return is smaller than the extra damage, and growth turns down. This effect is often called volatility drag.

It is also why the rule's line in the figure is still rising at three times the account where the hold line has collapsed. The rule was out of the market for the worst months. The worst month it actually held lost 13.5%, half the worst month of the fund. Smaller swings leave more room before size starts to destroy growth.

How much is too much: the Kelly idea

In 1956 John Kelly, a physicist working at Bell Labs, worked out the bet size that makes a gambler's money grow fastest in the long run when the odds are known. Bet less than that fraction and growth is slower but smoother. Bet more and growth is slower and rougher. Bet about twice the Kelly fraction and the long-run growth rate falls to zero.

The hold column of the table is that curve drawn from real data. Growth peaked somewhere between 1.5 and 2 times the account and then fell away. It is tempting to read the peak as the right size. Three things argue against it.

The peak is known only afterwards. The curve was drawn with every month of 2000 to 2026 already in hand. Kelly's result assumes the odds are known exactly. For a market they are estimated from a sample, and the estimate is wrong by an unknown amount.

The peak sits beside the cliff. Being wrong on the high side is far more costly than being wrong on the low side. At 1.5 times the hold line earned 9.8%; at three times, 3.6%; at four, nothing. A person who guessed too low gave up a little growth. A person who guessed too high lost the account.

The sample's worst month is not the worst possible month. The rule's column looks as if it could carry four times the account, because the worst month it held lost 13.5%. One held month of −25%, which the fund itself has produced, would have taken a four-times account to zero. A size chosen to fit the worst loss in a backtest is fitted to luck.

For these reasons people who use the Kelly idea in practice usually take a fraction of the calculated size, commonly half, and those who use no formula at all get much the same protection from the plain rule of keeping size at or below the full account.

Sizing by risk per trade

The second method works trade by trade. Choose the share of the account that may be lost if the trade fails, find the distance from the entry to the stop, and divide.

The point for a strategy is what this does across many trades. A wide stop gives a small position and a tight stop a large one, so each losing trade costs about the same share of the account. A run of ten straight losses at 1% each takes the account down 9.6%, not the 30% or 50% that ten full-size losers could. The lesson Position sizing: how many shares to buy for a set risk works through the method and its traps, including the gap that jumps past a stop.

Sizing by volatility

The third method changes size through time. When the market's recent swings are large, hold less. When they are small, hold more, up to some limit.

The test, on the same QQQ data, with no timing rule at all:

  1. At each month-end, measure the volatility of the last 12 monthly returns and express it per year.
  2. Set next month's position to 15% divided by that figure, capped at the full account. Recent volatility of 30% gives a half-size position; 15% or less gives full size.
  3. Hold the rest in cash.

The first 12 months are needed for the first measurement, so this test runs from January 2001 to October 2026, 310 months. Costs, interest and dividends are left out as before.

QQQ monthly, 2001–2026: size scaled to 15% volatility against full size, no costs $0.5 $1 $2 $5 $10 Growth of $1 Full size (buy and hold) Volatility-scaled size 0% −20% −40% −60% Fall from the peak 2005 2010 2015 2020 2025
QQQ monthly, 2001 to 2026: a position scaled to 15% volatility against holding at full size. No costs, interest or dividends.
Measure Volatility-scaled Full size Constant 80% size
Return per year 9.6% 9.7% 8.1%
Deepest fall −37.6% −67.8% −58.4%
Volatility of the curve 14.9% 21.6% not computed
Average position 80% 100% 80%

The scaled position averaged 80% of the account. It was at full size in 105 of the 310 months and as low as 28% in November 2001. In September 2002 it stood at 40%, in October 2008 at 54%, in June 2022 at 66%.

The third column is the fair comparison. Holding a constant 80% has the same average exposure and gave 8.1% a year with a deepest fall of 58.4%. Scaling by volatility did better on both lines with the same average size, because it held less in the stretches where most of the damage was done. Volatility tends to cluster: a violent month is more often followed by another than by a calm one, which is what gives the method something to work with. Moreira and Muir (2017) document the same effect for the US stock market and for several other return factors.

Where sizing rules fail

  • Sudden falls from calm. Volatility scaling looks backward. After a quiet 2019 the position was at full size at the end of January 2020 and still at 92% at the end of February, going into one of the fastest falls on record. The method helps in long, grinding declines and does little against a crash out of a clear sky.
  • Cutting size at the bottom. By the time recent volatility is high, much of the fall has happened. The scaled position was smallest in late 2001 and 2002, and so took a smaller part in the first leg of the recovery.
  • Turnover. Resetting the size every month trades every month. With real costs, and taxes on each sale in a taxable account, the advantage in the table shrinks. A common fix is to change size only when the target moves by more than some band.
  • Borrowing. Sizes above the full account need margin. Interest is charged on the loan, the broker can demand more collateral after a fall, and positions can be sold without notice if it is not supplied (Buying on margin: how margin loans, interest and leverage work). The table charges none of this.
  • One sample. All of the above is one fund over 27 years. The level of 15% and the 12-month window were chosen as round numbers and not tuned; other choices give different tables.

The methods compared

Method Size is set by Main strength Main weakness
Fixed share of account A constant fraction Simple; scales return and fall together Ignores how risky the market currently is
Fixed risk per trade Stop distance and a loss budget Every loss costs about the same A gap through the stop breaks the budget
Volatility target Recent swings of the market Smaller falls in long declines Late in sudden crashes; more trading
Kelly or a fraction of it Estimated odds and payoffs Names the size beyond which growth falls Needs odds nobody knows; errors are costly

How Finabase does it

Finabase's testing tools fix the size for you, which is worth knowing when reading their results.

  • Backtest buys or shorts a fixed $10,000 on every signal, with no trading cost. Size never grows with the account and never shrinks after losses, so its Gain is a sum of dollar results, not a compounded return, and it cannot show what a different size would have done. Gain / max cost divides the gain by the most capital tied up at one time.
  • Strategy lets you set an Amount to invest and give weights to sibling branches of an if/then rule, so "60% in one fund and 40% in another when the condition holds" can be replayed over two years. Weights are fixed shares of the account; there is no setting that scales them to volatility, and nothing above 100%.
  • Paper trading leaves size entirely to you: the account starts with $10,000 of virtual cash and you type the number of shares on each order (Practise with Finabase paper trading on the web). It is a good place to practise turning a risk budget and a stop into a share count before real money depends on it.
  • The chart draws the Average True Range, the usual measure of a symbol's daily swing, for setting a stop distance (Average True Range (ATR): volatility, stops and position size).

None of these tools suggests a size or a trade.

Frequently asked questions

Does position size change a strategy's win rate?

No. The same signals win and lose on the same trades at any size. Size changes how much each win and loss moves the account, and therefore the return, the deepest fall and whether the account survives a bad run.

What is the Kelly criterion in simple terms?

It is the bet size that makes money grow fastest over many repeated bets when the odds are known. Betting more than that amount lowers long-run growth and raises the swings. Because market odds are only estimated, people who use it usually take half or less of the calculated size.

Is leverage always a bad idea in a strategy?

It is not always bad, but it changes the arithmetic. In the QQQ example twice the size raised the buy-and-hold return from 7.9% to 9.9% a year and the deepest fall from 81.1% to 98.1%, before any interest on the loan. Losses grow in step with size while compound growth does not.

What is volatility targeting?

It is a sizing rule that holds a larger position when recent price swings are small and a smaller one when they are large, aiming to keep the account's own swings near a chosen level. In the test here a 15% target roughly halved the deepest fall of holding QQQ with almost no change in return, before costs.

Summary

Position size is the second half of every strategy and the half that decides how it feels to follow. With one fixed set of QQQ signals, sizes up to the full account traded return against the deepest fall almost one for one. Beyond the full account the falls kept growing while the return bent over, and for buy and hold it peaked near twice the account and reached zero at four times, because large losses need disproportionate gains to repair. The size that maximized growth can only be seen in hindsight and lies next to ruin, which is why practical sizing stays well below it. Scaling size to recent volatility reduced the worst fall from 67.8% to 37.6% at nearly the same return in this sample, at the price of more trading and no protection against a crash that begins in calm conditions.

Check your understanding

4 questions on this lesson. Each answer comes with the reason, and nothing is scored.

  1. Two people follow the same signals on the same fund. One commits half the account each time, the other all of it. What is the same for both?

    1. The deepest fall of the account
    2. The win rate
    3. The yearly return
    Show the answer

    B. The win rate The same trades win and lose at any size, so the win rate is identical. Size changes how far each trade moves the account: in the QQQ test half size had about half the return (4.2% against 8.0%) and about half the deepest fall (19.8% against 36.5%).

  2. Holding QQQ from 2000 earned 7.9% a year at full size, 9.9% at twice the size and 3.6% at three times, before interest. Why did tripling the size earn less?

    1. Larger positions pay higher commissions
    2. The signals get worse at larger sizes
    3. Large losses need disproportionately large gains to recover, so bigger swings eat compound growth
    Show the answer

    C. Large losses need disproportionately large gains to recover, so bigger swings eat compound growth A 26.2% monthly loss needs a 35.5% gain to repair; at three times the size it is a 78.6% loss that needs a gain of several hundred percent. Average return rises in step with size, but the damage from swings rises faster.

  3. A backtest shows a rule would have grown fastest at four times the account, and the worst month the rule ever held lost 13.5%. What is the main flaw in using that size?

    1. It fits the size to the worst loss in one sample; a single held month of −25% would wipe the account out
    2. Four is not a round fraction of the Kelly size
    3. The rule would trade more often at a larger size
    Show the answer

    A. It fits the size to the worst loss in one sample; a single held month of −25% would wipe the account out The growth-maximizing size is known only in hindsight and sits next to ruin. The fund itself lost 26.2% in one month of the same sample. Guessing too high costs the account; guessing too low costs a little growth.

  4. A volatility-target rule held 92% of the account in QQQ going into March 2020. Why so much, just before a crash?

    1. The rule forecasts returns and expected a rise
    2. It measures past swings, and 2019 had been calm
    3. The cap on position size had been removed
    Show the answer

    B. It measures past swings, and 2019 had been calm Volatility scaling looks backward. It cuts size after the market has become violent, which helps in long declines such as 2001–2002 and 2008, and does little against a sudden fall that starts in quiet conditions.

Exercise

One task to do in Finabase. Nothing is graded or saved.

Open the QQQ chart with the Average True Range and read the latest ATR (14) value and the latest price. Size a pretend trade the fixed-risk way for a $10,000 paper account: risk 1%, so $100, and for this exercise treat one ATR as the stop distance. Shares = $100 ÷ ATR. How many shares is that, how many dollars, and what share of the account? What happens to the share count if the ATR doubles?

Open the QQQ chart with the ATR
What you should find

A wider stop gives fewer shares: if the ATR doubles, the share count halves and the loss at the stop stays near $100. That is the point of the method. Check the dollar size too. If it comes out above $10,000, the stop is too tight for that risk budget without borrowing, and the lesson shows that sizes above the full account are where the damage starts. A gap past the stop can still lose more than $100.

Sources and further reading

  1. J. L. Kelly Jr., “A New Interpretation of Information Rate,” Bell System Technical Journal 35(4), 1956
  2. Edward O. Thorp, “The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market,” Handbook of Asset and Liability Management, vol. 1 (2006)
  3. Alan Moreira and Tyler Muir, “Volatility-Managed Portfolios,” Journal of Finance 72(4), 2017
  4. FINRA — Understanding margin accounts
  5. Finabase chart data — QQQ monthly bars, April 1999 to October 2026; sizes and results computed by Finabase