Lab: Same Strategy, Three Position-Sizing Models

What this lab establishes

Three backtests. Identical entry signals, identical exits, identical costs, identical stop, identical universe, identical date range. The only thing that differs is how large each position is.

By the end you will have three equity curves that look nothing like each other, produced by the same trading rules — and a clear answer to the question of how much of a system’s result comes from its entries.

Budget an hour: about fifteen minutes to run, and forty-five to look properly at what came back.

Prerequisites

• Risk per trade and stop distance
• Volatility-based position sizing
• Position sizing with SetPositionSize()

The shared signal set

Complete runnable AFL

lab-shared-signals.afl

// lab-shared-signals.afl
// Part 34 - Lab: Same Strategy, Three Position-Sizing Models
//
// The reference copy of everything the three sizing variants have in common:
// the account, the execution assumptions, the costs, the universe filter, the
// entry and exit rules, the ranking rule and the stop. Nothing here decides how
// large a position is.
//
// Run this file as a SCAN to inspect the signals themselves. Do NOT backtest
// it: with no position size set, the backtester puts 100% of the account into
// the first symbol that signals and the result is meaningless. The three
// sizing variants are the runnable backtests.
//
// The block between the two marker lines below is copied verbatim into
// lab-sizing-fixed-shares.afl, lab-sizing-percent-equity.afl and
// lab-sizing-atr-risk.afl. Compare them before you run anything: if the blocks
// have drifted apart, the comparison is measuring the wrong thing.

// ===== SHARED BLOCK BEGIN - byte-identical in all four lab files ============
//
// ASSUMPTIONS - the same in every variant, so that the comparison is fair:
//   - Daily bars, split- and dividend-adjusted end-of-day data.
//   - Signals are read on the close of the signal bar. Every order fills at
//     the NEXT bar's open. SetTradeDelays(1,1,1,1) enforces that.
//   - Commission 0.1% of trade value, charged on entry and on exit.
//   - Slippage is NOT modelled. Every variant is flattered by the same amount.
//   - ExitAtStop = 1: stops are checked against the bar's High-Low range and
//     filled at the stop level. Optimistic, and optimistic identically in all
//     three variants, which is what keeps the comparison honest.
//   - Whole shares, one currency, no margin, long only.
//   - Starting capital 100,000 in whatever currency your database is in.

// ---- Account and execution ------------------------------------------------
MaxPositions = 10;

SetOption( "InitialEquity", 100000 );
SetOption( "MaxOpenPositions", MaxPositions );
SetOption( "AllowPositionShrinking", True );
SetOption( "CommissionMode", 1 );        // 1 = percent of trade value
SetOption( "CommissionAmount", 0.1 );    // 0.1% each way
SetOption( "ActivateStopsImmediately", True );

SetTradeDelays( 1, 1, 1, 1 );
BuyPrice  = Open;
SellPrice = Open;
RoundLotSize = 1;

// ---- Universe filter ------------------------------------------------------
MinTurnover = 2000000;
Liquid = MA( Close * Volume, 50 ) > MinTurnover;

// ---- Signals --------------------------------------------------------------
TrendPeriod = 200;
EntryPeriod = 50;

Trend = Close > MA( Close, TrendPeriod );
Buy   = Cross( Close, MA( Close, EntryPeriod ) ) AND Trend AND Liquid;
Sell  = Cross( MA( Close, EntryPeriod ), Close );

// Ranking, used when more symbols signal on one bar than there are free slots.
PositionScore = 100 - RSI( 14 );

// ---- The stop -------------------------------------------------------------
// Three ATRs below the entry price, sampled on the signal bar and held for the
// life of the trade. Identical in all three variants: the exit is not what is
// being compared.
AtrPeriod    = 20;
StopAtrMult  = 3.0;
MinAtrPct    = 0.5;

UsableAtr    = Max( ATR( AtrPeriod ), Close * MinAtrPct / 100 );
StopDistance = StopAtrMult * UsableAtr;

// SetTradeDelays shifts Buy/Sell/Short/Cover and nothing else, so the stop
// amount is shifted here by hand to make the value read on the entry bar the
// one computed on the signal bar.
StopAtEntry = Nz( Ref( StopDistance, -1 ), 1 );
ApplyStop( stopTypeLoss, stopModePoint, StopAtEntry, 1 );
//
// ===== SHARED BLOCK END =====================================================

Download lab-shared-signals.afl80 lines

This is the reference copy of everything the three variants have in common: the account, the execution assumptions, the costs, the universe filter, the entry and exit rules, the ranking rule and the stop. Nothing in it decides how large a position is.

⚠Warning

Run this file as a Scan to inspect the signals. Do not backtest it. With no position size set, the backtester puts 100% of the account into the first symbol that signals, and the result is meaningless. The three variants below are the runnable backtests.

Two details in the shared block that are easy to get wrong

The stop amount is shifted by hand.

Fragment — not a complete formula

StopAtEntry = Nz( Ref( StopDistance, -1 ), 1 );
ApplyStop( stopTypeLoss, stopModePoint, StopAtEntry, 1 );

SetTradeDelays() shifts Buy, Sell, Short and Cover — and nothing else. So an array computed on the signal bar is still lined up with the undelayed signal. Shifting it by hand makes the value read on the entry bar the one that was computed on the signal bar, which is what you meant.

The ATR has a floor.

Fragment — not a complete formula

UsableAtr = Max( ATR( AtrPeriod ), Close * MinAtrPct / 100 );

Without it, a symbol whose ATR has collapsed to nearly zero produces a stop distance of nearly zero — and in model 3, a position size of nearly infinity. The floor is a modelling decision, and it is the kind that has to be made before the run rather than after it goes wrong.

!Important

Two assumptions in the shared block are deliberately optimistic, and both are optimistic identically in all three variants, which is what keeps the comparison honest:

• Slippage is not modelled at all.
• ExitAtStop = 1 fills every stop at exactly the nominated level, including on bars that gapped through it — see the stops lesson.

The absolute numbers from this lab are therefore all too good. The comparison between them is what the lab is for, and it is unaffected.

Model 1: fixed shares

Complete runnable AFL

lab-sizing-fixed-shares.afl

// lab-sizing-fixed-shares.afl
// Part 34 - Lab: Same Strategy, Three Position-Sizing Models
// SIZING MODEL 1 OF 3: A FIXED NUMBER OF SHARES
//
// Buy the same share count every time, whatever the share costs and whatever
// the account is worth. The oldest sizing rule there is, and the one that puts
// wildly different amounts of money and wildly different amounts of risk into
// each position without ever saying so.
//
// Everything except the four lines under "SIZING MODEL 1" is shared with the
// other two variants and with lab-shared-signals.afl.

// ===== SHARED BLOCK BEGIN - byte-identical in all four lab files ============
//
// ASSUMPTIONS - the same in every variant, so that the comparison is fair:
//   - Daily bars, split- and dividend-adjusted end-of-day data.
//   - Signals are read on the close of the signal bar. Every order fills at
//     the NEXT bar's open. SetTradeDelays(1,1,1,1) enforces that.
//   - Commission 0.1% of trade value, charged on entry and on exit.
//   - Slippage is NOT modelled. Every variant is flattered by the same amount.
//   - ExitAtStop = 1: stops are checked against the bar's High-Low range and
//     filled at the stop level. Optimistic, and optimistic identically in all
//     three variants, which is what keeps the comparison honest.
//   - Whole shares, one currency, no margin, long only.
//   - Starting capital 100,000 in whatever currency your database is in.

// ---- Account and execution ------------------------------------------------
MaxPositions = 10;

SetOption( "InitialEquity", 100000 );
SetOption( "MaxOpenPositions", MaxPositions );
SetOption( "AllowPositionShrinking", True );
SetOption( "CommissionMode", 1 );        // 1 = percent of trade value
SetOption( "CommissionAmount", 0.1 );    // 0.1% each way
SetOption( "ActivateStopsImmediately", True );

SetTradeDelays( 1, 1, 1, 1 );
BuyPrice  = Open;
SellPrice = Open;
RoundLotSize = 1;

// ---- Universe filter ------------------------------------------------------
MinTurnover = 2000000;
Liquid = MA( Close * Volume, 50 ) > MinTurnover;

// ---- Signals --------------------------------------------------------------
TrendPeriod = 200;
EntryPeriod = 50;

Trend = Close > MA( Close, TrendPeriod );
Buy   = Cross( Close, MA( Close, EntryPeriod ) ) AND Trend AND Liquid;
Sell  = Cross( MA( Close, EntryPeriod ), Close );

// Ranking, used when more symbols signal on one bar than there are free slots.
PositionScore = 100 - RSI( 14 );

// ---- The stop -------------------------------------------------------------
// Three ATRs below the entry price, sampled on the signal bar and held for the
// life of the trade. Identical in all three variants: the exit is not what is
// being compared.
AtrPeriod    = 20;
StopAtrMult  = 3.0;
MinAtrPct    = 0.5;

UsableAtr    = Max( ATR( AtrPeriod ), Close * MinAtrPct / 100 );
StopDistance = StopAtrMult * UsableAtr;

// SetTradeDelays shifts Buy/Sell/Short/Cover and nothing else, so the stop
// amount is shifted here by hand to make the value read on the entry bar the
// one computed on the signal bar.
StopAtEntry = Nz( Ref( StopDistance, -1 ), 1 );
ApplyStop( stopTypeLoss, stopModePoint, StopAtEntry, 1 );
//
// ===== SHARED BLOCK END =====================================================

// ---- SIZING MODEL 1: FIXED SHARES ------------------------------------------
// 200 shares is 4,000 in a share priced at 20 and 100,000 - the entire account -
// in a share priced at 500. The rule does not know the difference. Watch the
// Detailed log to see how often "position size shrinking" rescues it.
FixedShares = 200;

SetPositionSize( FixedShares, spsShares );

Download lab-sizing-fixed-shares.afl82 lines

Fragment — not a complete formula

FixedShares = 200;
SetPositionSize( FixedShares, spsShares );

Buy the same share count every time, whatever the share costs and whatever the account is worth. The oldest sizing rule there is.

What it actually does: 200 shares is 4,000 currency units in a share priced at 20, and 100,000 — the entire account — in a share priced at 500. The rule has no idea. It puts wildly different amounts of money, and wildly different amounts of risk, into each position without ever saying so.

What to watch for: how often AllowPositionShrinking rescues it. With that option on, a position too large for the remaining cash is reduced rather than skipped, which means the effective sizing rule is “200 shares, or whatever I can afford, whichever is smaller” — and that is a rule nobody chose.

Model 2: percent of equity

Complete runnable AFL

lab-sizing-percent-equity.afl

// lab-sizing-percent-equity.afl
// Part 34 - Lab: Same Strategy, Three Position-Sizing Models
// SIZING MODEL 2 OF 3: A FIXED PERCENTAGE OF PORTFOLIO EQUITY
//
// Every position is the same fraction of the account: a tenth of equity, ten
// positions, fully invested when the book is full. Equal money, deliberately
// unequal risk - a tenth of equity in a placid utility and a tenth in a
// small-cap miner are not the same bet.
//
// Everything except the three lines under "SIZING MODEL 2" is shared with the
// other two variants and with lab-shared-signals.afl.

// ===== SHARED BLOCK BEGIN - byte-identical in all four lab files ============
//
// ASSUMPTIONS - the same in every variant, so that the comparison is fair:
//   - Daily bars, split- and dividend-adjusted end-of-day data.
//   - Signals are read on the close of the signal bar. Every order fills at
//     the NEXT bar's open. SetTradeDelays(1,1,1,1) enforces that.
//   - Commission 0.1% of trade value, charged on entry and on exit.
//   - Slippage is NOT modelled. Every variant is flattered by the same amount.
//   - ExitAtStop = 1: stops are checked against the bar's High-Low range and
//     filled at the stop level. Optimistic, and optimistic identically in all
//     three variants, which is what keeps the comparison honest.
//   - Whole shares, one currency, no margin, long only.
//   - Starting capital 100,000 in whatever currency your database is in.

// ---- Account and execution ------------------------------------------------
MaxPositions = 10;

SetOption( "InitialEquity", 100000 );
SetOption( "MaxOpenPositions", MaxPositions );
SetOption( "AllowPositionShrinking", True );
SetOption( "CommissionMode", 1 );        // 1 = percent of trade value
SetOption( "CommissionAmount", 0.1 );    // 0.1% each way
SetOption( "ActivateStopsImmediately", True );

SetTradeDelays( 1, 1, 1, 1 );
BuyPrice  = Open;
SellPrice = Open;
RoundLotSize = 1;

// ---- Universe filter ------------------------------------------------------
MinTurnover = 2000000;
Liquid = MA( Close * Volume, 50 ) > MinTurnover;

// ---- Signals --------------------------------------------------------------
TrendPeriod = 200;
EntryPeriod = 50;

Trend = Close > MA( Close, TrendPeriod );
Buy   = Cross( Close, MA( Close, EntryPeriod ) ) AND Trend AND Liquid;
Sell  = Cross( MA( Close, EntryPeriod ), Close );

// Ranking, used when more symbols signal on one bar than there are free slots.
PositionScore = 100 - RSI( 14 );

// ---- The stop -------------------------------------------------------------
// Three ATRs below the entry price, sampled on the signal bar and held for the
// life of the trade. Identical in all three variants: the exit is not what is
// being compared.
AtrPeriod    = 20;
StopAtrMult  = 3.0;
MinAtrPct    = 0.5;

UsableAtr    = Max( ATR( AtrPeriod ), Close * MinAtrPct / 100 );
StopDistance = StopAtrMult * UsableAtr;

// SetTradeDelays shifts Buy/Sell/Short/Cover and nothing else, so the stop
// amount is shifted here by hand to make the value read on the entry bar the
// one computed on the signal bar.
StopAtEntry = Nz( Ref( StopDistance, -1 ), 1 );
ApplyStop( stopTypeLoss, stopModePoint, StopAtEntry, 1 );
//
// ===== SHARED BLOCK END =====================================================

// ---- SIZING MODEL 2: PERCENT OF EQUITY -------------------------------------
// 100 / MaxPositions, so a full book is fully invested. Because the percentage
// is applied to CURRENT portfolio equity, this model compounds: positions grow
// after a good run and shrink after a bad one, without any rule saying so.

SetPositionSize( 100 / MaxPositions, spsPercentOfEquity );

Download lab-sizing-percent-equity.afl81 lines

Fragment — not a complete formula

SetPositionSize( 100 / MaxPositions, spsPercentOfEquity );

Every position is the same fraction of the account: a tenth of equity, ten positions, fully invested when the book is full.

Equal money, deliberately unequal risk. A tenth of equity in a placid utility and a tenth in a small-cap miner are not the same bet — the second can move three times as far in a day.

It compounds. Because the percentage applies to current portfolio equity, positions grow after a good run and shrink after a bad one, with no rule saying so. That is usually desirable, and it is also what makes the participation problem from Part 30 grow with the backtest.

Model 3: ATR risk-based

Complete runnable AFL

lab-sizing-atr-risk.afl

// lab-sizing-atr-risk.afl
// Part 34 - Lab: Same Strategy, Three Position-Sizing Models
// SIZING MODEL 3 OF 3: RISK-BASED SIZING FROM THE ATR STOP
//
// Size each position so that the distance to the stop - which the shared block
// already defined as three ATRs - costs the same percentage of equity in every
// symbol. Equal risk, deliberately unequal money.
//
// Everything except the block under "SIZING MODEL 3" is shared with the other
// two variants and with lab-shared-signals.afl.

// ===== SHARED BLOCK BEGIN - byte-identical in all four lab files ============
//
// ASSUMPTIONS - the same in every variant, so that the comparison is fair:
//   - Daily bars, split- and dividend-adjusted end-of-day data.
//   - Signals are read on the close of the signal bar. Every order fills at
//     the NEXT bar's open. SetTradeDelays(1,1,1,1) enforces that.
//   - Commission 0.1% of trade value, charged on entry and on exit.
//   - Slippage is NOT modelled. Every variant is flattered by the same amount.
//   - ExitAtStop = 1: stops are checked against the bar's High-Low range and
//     filled at the stop level. Optimistic, and optimistic identically in all
//     three variants, which is what keeps the comparison honest.
//   - Whole shares, one currency, no margin, long only.
//   - Starting capital 100,000 in whatever currency your database is in.

// ---- Account and execution ------------------------------------------------
MaxPositions = 10;

SetOption( "InitialEquity", 100000 );
SetOption( "MaxOpenPositions", MaxPositions );
SetOption( "AllowPositionShrinking", True );
SetOption( "CommissionMode", 1 );        // 1 = percent of trade value
SetOption( "CommissionAmount", 0.1 );    // 0.1% each way
SetOption( "ActivateStopsImmediately", True );

SetTradeDelays( 1, 1, 1, 1 );
BuyPrice  = Open;
SellPrice = Open;
RoundLotSize = 1;

// ---- Universe filter ------------------------------------------------------
MinTurnover = 2000000;
Liquid = MA( Close * Volume, 50 ) > MinTurnover;

// ---- Signals --------------------------------------------------------------
TrendPeriod = 200;
EntryPeriod = 50;

Trend = Close > MA( Close, TrendPeriod );
Buy   = Cross( Close, MA( Close, EntryPeriod ) ) AND Trend AND Liquid;
Sell  = Cross( MA( Close, EntryPeriod ), Close );

// Ranking, used when more symbols signal on one bar than there are free slots.
PositionScore = 100 - RSI( 14 );

// ---- The stop -------------------------------------------------------------
// Three ATRs below the entry price, sampled on the signal bar and held for the
// life of the trade. Identical in all three variants: the exit is not what is
// being compared.
AtrPeriod    = 20;
StopAtrMult  = 3.0;
MinAtrPct    = 0.5;

UsableAtr    = Max( ATR( AtrPeriod ), Close * MinAtrPct / 100 );
StopDistance = StopAtrMult * UsableAtr;

// SetTradeDelays shifts Buy/Sell/Short/Cover and nothing else, so the stop
// amount is shifted here by hand to make the value read on the entry bar the
// one computed on the signal bar.
StopAtEntry = Nz( Ref( StopDistance, -1 ), 1 );
ApplyStop( stopTypeLoss, stopModePoint, StopAtEntry, 1 );
//
// ===== SHARED BLOCK END =====================================================

// ---- SIZING MODEL 3: ATR RISK-BASED ----------------------------------------
// Position value / equity = RiskPercent * Price / StopDistance, so that a move
// of StopDistance against the position costs RiskPercent of equity.
//
// The cap matters. Without it, a symbol whose ATR has collapsed asks for a
// position several times the size of the account, and "allow position size
// shrinking" quietly turns the risk rule into a bet-the-farm rule.
RiskPercent    = 1.0;
MaxSizePercent = 20;

SizePercent = RiskPercent * Close / StopDistance;
SizePercent = Min( SizePercent, MaxSizePercent );

// Shifted by the buy delay for the same reason the stop was.
SizeAtEntry = Nz( Ref( SizePercent, -1 ), 0 );

SetPositionSize( SizeAtEntry, spsPercentOfEquity );

Download lab-sizing-atr-risk.afl91 lines

Fragment — not a complete formula

RiskPercent    = 1.0;
MaxSizePercent = 20;

SizePercent = RiskPercent * Close / StopDistance;
SizePercent = Min( SizePercent, MaxSizePercent );

SizeAtEntry = Nz( Ref( SizePercent, -1 ), 0 );
SetPositionSize( SizeAtEntry, spsPercentOfEquity );

Size each position so that the distance to the stop — three ATRs, defined in the shared block — costs the same percentage of equity in every symbol.

The arithmetic. You want a move of StopDistance against the position to cost RiskPercent of equity. A position worth V loses V × StopDistance / Price when price falls by StopDistance. Setting that equal to RiskPercent × Equity / 100 and solving for V / Equity gives RiskPercent × Price / StopDistance, expressed as a percentage. That is the line.

Equal risk, deliberately unequal money. A volatile symbol gets a small position and a placid one gets a large position, so that a stop-out costs the same either way.

✕Common mistake

The cap is not optional. Without MaxSizePercent, a symbol whose ATR has collapsed asks for a position several times the size of the account — and AllowPositionShrinking then quietly turns your careful risk rule into a bet-the-farm rule by giving it everything available.

This is the most common failure of risk-based sizing, and the symptom is a backtest with a handful of enormous positions in unusually quiet instruments.

Running all three

1. Run each of the three variants over the same universe and the same date range.
2. Change nothing else between runs — not the settings dialog, not the watch list, not the report options.
3. Record the figures below from each report.

	Fixed shares	% of equity	ATR risk
Number of trades
Net Profit %
Annual Return %
Max. system % drawdown
CAR/MaxDD
Exposure %
Winners %
Largest single loss (from the trade list)
Largest position value (from the trade list)

=Expected result

What you should see, and why:

• Trade counts differ, even though the signals are identical. Sizing decides affordability, and affordability decides which signals were taken when candidates outnumbered slots or cash.
• The three equity curves have different shapes, not just different heights.
• Fixed shares produces the most erratic result. Its effective risk per trade varies by an order of magnitude across symbols, so a few positions dominate.
• ATR risk-based produces the most even distribution of losses. That is the whole point of it, and it is visible in the trade list before it is visible in any summary statistic.
• The drawdowns differ more than the returns do. This is the finding, and it is why the drawdown row is in bold.

What this proves about entry signals

Here is the conclusion the lab exists to deliver:

The entry signals were identical in all three runs. Same bars, same symbols, same conditions. And the three results are not variations on a theme — they are different systems, with different risk, different drawdowns and, quite possibly, different signs.

Two consequences follow, and both are uncomfortable if you have spent most of your effort on entries.

A backtest result is not a property of the rules. It is a property of the rules plus the sizing plus the constraints plus the costs. Quoting a return figure without the sizing model is quoting an incomplete specification.

Comparing two published systems’ returns is close to meaningless unless they used the same sizing model. Most of the difference between them may be sizing.

∴Research note

This does not mean sizing is more important than entries, and it is worth being precise about why. Sizing cannot create an edge that is not there: apply the best sizing model in the world to random entries and you will get a well-controlled random result.

What sizing does is determine how much of whatever edge exists survives the path — how large the drawdowns are, whether a bad run ends the account, and whether a few positions dominate. Entries decide whether there is anything to harvest; sizing decides whether you are still there to harvest it.

Verification checklist

Before you interpret anything, confirm the comparison is actually fair.

1. The shared block is byte-identical in all four files. Diff them. If they have drifted, the comparison is measuring the drift.
2. The date range and universe are identical. Check the Analysis window, not your memory.
3. The signal count is identical. Run the shared file as a Scan and compare the signal total against each variant’s “signals generated”. Entry rules that differ would invalidate everything.
4. The trade counts differ — they should. If they are identical, sizing is not binding and the lab has nothing to show you; increase the account size difference or reduce the slot count.
5. Hand-check one trade in each variant: same symbol, same entry date, same entry price, different share count. If the entry date differs, something other than sizing changed.
6. Look at the largest position in each trade list against the volume on its entry bar. Model 1 and model 3 can both produce positions nobody could have filled.

▷Exercise

Two extensions, in order of value.

Vary RiskPercent in model 3 across 0.5%, 1% and 2%, holding everything else fixed. Plot the three drawdowns against the three returns. The relationship is usually much less linear than people expect, and finding where it stops being worth it is the practical question this lab feeds into.

Remove MaxSizePercent from model 3 and re-run. Then find the largest position in the trade list and look at that symbol’s ATR on the entry bar. Seeing the failure once is worth more than the warning above.

Recap

Same signals, three sizing rules, three different systems. Fixed shares varies the money and the risk without saying so. Percent of equity equalises the money and leaves the risk unequal, and it compounds. ATR risk-based equalises the risk and leaves the money unequal — provided it is capped, because uncapped it converts a low-volatility symbol into a bet on the whole account. The drawdowns differ more than the returns do, and none of it changed a single entry signal. A backtest result is a property of the whole specification, not of the rules.

Check your understanding

Question 1. The three variants share identical entry signals, yet their trade counts differ. Why?

• The random seed differs between runs
• Sizing decides affordability, and affordability decides which signals were taken when candidates outnumbered slots or available cash
• ApplyStop behaves differently under different sizing models
• The trade count in AmiBroker includes scale-ins

Show the answer and why

Answer: Sizing decides affordability, and affordability decides which signals were taken when candidates outnumbered slots or available cash

The signals are the same; which of them could be acted on is not. A larger position size exhausts cash sooner, so fewer subsequent candidates are affordable — and in the default backtest mode, a skipped entry then silences that symbol until an exit arrives.

Question 2. Why is the stop distance shifted by hand with Ref( StopDistance, -1 ) when SetTradeDelays( 1, 1, 1, 1 ) is already in force?

StopAtEntry = Nz( Ref( StopDistance, -1 ), 1 );
ApplyStop( stopTypeLoss, stopModePoint, StopAtEntry, 1 );

• To make the stop wider
• Because trade delays shift only the Buy/Sell/Short/Cover arrays — any other array stays lined up with the undelayed signal
• Because ApplyStop cannot accept an unshifted array
• To avoid Null values at the start of the array

Show the answer and why

Answer: Because trade delays shift only the Buy/Sell/Short/Cover arrays — any other array stays lined up with the undelayed signal

The delay mechanism is documented as shifting the four signal arrays and nothing else. Without the manual shift, the stop amount read on the entry bar would be the one computed on the entry bar, not the one computed on the signal bar that justified the trade.

Question 3. What does MaxSizePercent protect against in the ATR risk-based model?

SizePercent = RiskPercent * Close / StopDistance;
SizePercent = Min( SizePercent, MaxSizePercent );

• Positions that are too small to be worth the commission
• A symbol whose ATR has collapsed producing a tiny stop distance and therefore an enormous requested position — which AllowPositionShrinking then fills with everything available
• Rounding errors in the share count
• Positions in symbols that are too illiquid

Show the answer and why

Answer: A symbol whose ATR has collapsed producing a tiny stop distance and therefore an enormous requested position — which AllowPositionShrinking then fills with everything available

Position size is inversely proportional to stop distance, so as the stop distance approaches zero the requested size approaches infinity. Position shrinking then hands it the whole account, converting a 1%-risk rule into a bet-the-farm rule. The ATR floor in the shared block is the same defence applied earlier.

Question 4. Which conclusions does this lab support? Select all that apply.

• A backtest result is a property of the rules plus the sizing plus the constraints plus the costs, not of the rules alone
• Comparing two published systems on return alone is close to meaningless if their sizing models differ
• Position sizing can create an edge where the entry rules have none
• The drawdowns differ between the models more than the returns do

Show the answer and why

Answer: A backtest result is a property of the rules plus the sizing plus the constraints plus the costs, not of the rules alone, Comparing two published systems on return alone is close to meaningless if their sizing models differ, The drawdowns differ between the models more than the returns do

Sizing determines how much of an existing edge survives the path — how deep the drawdowns are and whether a bad run ends the account. It cannot manufacture an edge: apply the best sizing model to random entries and you get a well-controlled random result.

Question 5. The lab uses ExitAtStop = 1 and no slippage. Does that invalidate its conclusions?

• Yes — the results are unusable
• No, because both assumptions are optimistic identically across all three variants, so the absolute numbers are too good but the comparison between them is unaffected
• No, because ExitAtStop = 1 is the correct setting for daily data
• Yes, unless the universe is restricted to large caps

Show the answer and why

Answer: No, because both assumptions are optimistic identically across all three variants, so the absolute numbers are too good but the comparison between them is unaffected

A shared optimistic assumption inflates all three results by a similar mechanism, so the differences — which is what the lab measures — survive. What you must not do is quote any single one of these absolute figures as a realistic result.

Check my answersClear