Project: Your First Complete Trading System

By the end of this session you will have a complete trading system: a written hypothesis, a rule set that passes the ambiguity test, an assumptions block, a formula that runs, a backtest report, and a page of notes recording what you concluded and what you refused to conclude.

You will also have resisted the strongest urge in this entire course. The system has two parameters. Both were chosen from convention before anything was run. Neither will be adjusted. The final section explains at length why that restraint is the most valuable thing in the project, but it is worth stating the short version now: you cannot recognise an overfitted result until you have seen an honest one produced by the same rules. This project manufactures that baseline.

What you are building

1. 1HypothesisWritten and dated first
2. 2RulesAmbiguity-tested
3. 3AssumptionsCosts, fills, liquidity
4. 4FormulaOne file, runnable
5. 5ReportRead in a fixed order
6. 6NotesWhat you may and may not say

What you need

An AmiBroker installation with a daily end-of-day database, and a watch list of at least 100 symbols with a reasonable spread of liquidity. Nothing in this project needs a real-time feed, a data subscription or the Professional edition. Allow an hour: about fifteen minutes of writing, ten of setup, five of running, and the rest reading and arguing with the result.

1. The hypothesis

Written before anything else, and quoted here in full so that later sections can be checked against it rather than against a memory of it.

Observation. Liquid shares that spend long stretches above a slow moving average appear to keep rising for a while after crossing it.

Claim. In a universe of liquid shares, a long position opened after the close crosses above its own 200-day simple moving average, and closed on the reverse cross or after 60 bars — whichever comes first — produces a positive average result per trade after the stated costs.

Mechanism. Slow, uneven diffusion of information, and the reluctance of holders to sell winners. Both are plausible; neither is established by this test, and this formula cannot distinguish between them.

Prior. Low. The rule is one of the most widely known in the field, which means any effect that was easy to capture has had decades of attention.

Falsified by. An average result per trade that is not positive after costs; or one that is positive only in a single sub-period; or one that disappears when the moving-average length is moved a little in either direction.

Cost assumption, decided now. 0.10% slippage per side plus 0.10% commission per side, so roughly 0.40% round trip.

Notice the prior. Being explicit that you expect this to be weak changes how you will read the report, and it makes a strong result more interesting rather than less.

2. The rules

Section	Specification
Universe	The watch list this is applied to, restricted to symbols whose 50-bar average turnover (close × volume) on the signal bar is at least 5,000,000 in the quote currency
Data	Your end-of-day database, split- and dividend-adjusted daily bars. Record the vendor and refresh date
Entry	On the close of bar t: the close crosses above its own 200-bar simple moving average, the liquidity floor is met, and one position slot would be no more than 1% of the symbol’s average turnover. Event, not state
Exit	Whichever comes first: the close crosses back below the 200-bar average, or 60 bars have elapsed since entry. No stop
Sizing	10 slots, 10% of portfolio equity each, on 100,000 initial equity, no margin
Execution	One bar of delay on all four signals; fill at the next bar’s open, adjusted by 0.10% against us; 0.10% commission per side
Selection	When more entries signal than there are free slots, prefer the most liquid candidate

Two entries in that table deserve comment because they look like omissions and are not.

No stop. A stop is a second rule with its own parameter, and adding one now would mean two things changed at once between “no system” and “a system”. It also means every loss in this test is bounded only by the exit rule and the 60-bar clock, which will show up in the report as larger individual losses than a stopped system would produce. That is information, not a defect. Parts 28 and 34 add stops properly, and you will then be able to measure what a stop actually did, because you have this run to compare against.

A stated tie-break. When eleven symbols signal and three slots are free, something decides. Left unspecified, that something is invisible to a reader and hard to reproduce. Preferring the most liquid candidate is a choice made on grounds of fill realism rather than expected return, which keeps it out of the business of improving the result.

∴Research note

Run the solo ambiguity test on the rules table before you write any AFL. Close this page, write the rules again from memory, and compare. The two places readers most often diverge here are whether “crosses above” means an event or a state, and whether the liquidity floor is checked on the signal bar or the fill bar. Both are specified above; check that your version specifies them too.

3. The assumptions block and the formula

The formula reuses the harness from the previous lesson: a header a reader can evaluate without AmiBroker, then constants, then settings, then the rules.

Complete runnable AFL

first-system.afl

// first-system.afl
// Part 27 - Project: Your First Complete Trading System
//
// A deliberately plain long-only trend system, written so that every decision
// behind it is visible on the page. It has two numbers in it and neither of
// them has been searched for. That is the point of the exercise: before you can
// recognise a result that has been fitted to its data, you need to have seen
// one that has not.
//
// ============================ HYPOTHESIS ==============================
//   Observation  Liquid shares that spend long stretches above a slow moving
//                average appear to keep rising for a while after crossing it.
//   Claim        In a universe of liquid shares, a long position opened after
//                the close crosses above its own 200-day simple moving average
//                and closed on the reverse cross or after 60 bars, whichever
//                comes first, produces a positive average result per trade
//                after the costs stated below.
//   Mechanism    Slow, uneven diffusion of information and the reluctance of
//                holders to sell winners. Both are documented behaviours in the
//                academic literature and both are plausible here; neither is
//                established by this test, and this formula cannot distinguish
//                between them.
//   Falsified by An average result per trade that is not positive after costs,
//                or one that is positive only in a single sub-period, or one
//                that disappears when the moving-average length is moved a
//                little in either direction.
// ======================================================================
//
// ============================ ASSUMPTIONS =============================
//   Data source     Your end-of-day database. Record the vendor and the date
//                   the history was last refreshed alongside the result.
//   Adjustment      Split- and dividend-adjusted daily bars.
//   Interval        Daily.
//   Universe        The watch list this is applied to, further restricted by
//                   the liquidity floor below.
//   Survivorship    NOT controlled. If the watch list is today's index members
//                   the answer is about survivors, which is a different and
//                   easier question. See Part 30.
//   Signal timing   Computed on the close of bar t from data up to bar t.
//   Fill            The opening print of bar t+1, in full.
//   Trade delays    1 bar on buy, sell, short and cover.
//   Slippage        0.10% per side, applied by moving BuyPrice and SellPrice.
//   Commission      0.10% of trade value per side, CommissionMode 1.
//   Liquidity floor 50-day average turnover of at least 5,000,000 in the quote
//                   currency on the signal bar.
//   Participation   One slot capped at 1% of that average turnover.
//   Position sizing 10 slots, 10% of portfolio equity each.
//   Initial equity  100,000 in the quote currency.
//   Interest        0% on idle cash.
//   Selection       When more entry signals appear than there are free slots,
//                   the most liquid candidate is preferred. Recorded here
//                   because otherwise the choice would be invisible.
//   Stops           NONE. This is a decision, not an oversight: a stop is a
//                   second rule with its own parameter, and adding one now
//                   would mean two things changed at once. Parts 28 and 34
//                   add stops properly. Until then, the only thing limiting a
//                   losing trade is the exit rule and the 60-bar clock.
//   Short selling   Not simulated.
//   NOT MODELLED    Bid-ask spread beyond the slippage figure; market impact
//                   beyond the participation cap; taxes; borrow costs;
//                   partial fills; corporate actions your data may not carry.
// ======================================================================
//
// How to run it:
//   Formula Editor -> paste -> Tools -> Send to Analysis
//   Apply to: Filter, and choose a watch list of at least 100 liquid symbols
//   Range:    From-To dates, chosen and written down BEFORE you press Backtest
//   Settings: Periodicity = Daily
//   Then press Backtest, then Report.

// ------------------------------------------------ the two rule numbers
// Chosen once, from convention, and left alone. 200 because it is the length
// most widely quoted, which makes it the length least likely to have been
// picked because it flattered this particular data. 60 because it is about a
// quarter of a trading year and bounds how long capital sits in one idea.
MaPeriod    = 200;
MaxHoldBars = 60;

// ---------------------------------------------------------------- costs
SlippagePct   = 0.10;
CommissionPct = 0.10;

// ------------------------------------------------------------ portfolio
StartEquity   = 100000;
PositionSlots = 10;
PositionPct   = 100 / PositionSlots;

// ------------------------------------------------------------ liquidity
TurnoverWindow   = 50;
MinTurnover      = 5000000;
MaxParticipation = 0.01;

SetOption( "InitialEquity",    StartEquity );
SetOption( "MaxOpenPositions", PositionSlots );
SetOption( "AccountMargin",    100 );
SetOption( "InterestRate",     0 );
SetOption( "CommissionMode",   1 );
SetOption( "CommissionAmount", CommissionPct );

SetPositionSize( PositionPct, spsPercentOfEquity );
SetTradeDelays( 1, 1, 1, 1 );

BuyPrice  = Open * ( 1 + SlippagePct / 100 );
SellPrice = Open * ( 1 - SlippagePct / 100 );

// ------------------------------------------------------------ the rules
TrendLine = MA( Close, MaPeriod );

AverageTurnover = MA( Close * Volume, TurnoverWindow );
LiquidEnough    = AverageTurnover >= MinTurnover;

SlotValue   = StartEquity * PositionPct / 100;
SmallEnough = SlotValue <= MaxParticipation * AverageTurnover;

// Both filters are applied to the entry only. An exit must never be filtered
// away: if a symbol dries up while we hold it, we still want out.
Buy  = Cross( Close, TrendLine ) AND LiquidEnough AND SmallEnough;
Sell = Cross( TrendLine, Close );

Buy  = ExRem( Buy,  Sell );
Sell = ExRem( Sell, Buy  );

// The time exit. The N-bar stop is counted by the backtester from the entry
// bar, which is why it is expressed here rather than with BarsSince() on a
// signal array that the backtester is about to shift by one bar anyway.
// SetTradeDelays only shifts Buy/Sell/Short/Cover; an exit produced by
// ApplyStop is not one of those four arrays and is not shifted.
ApplyStop( stopTypeNBar, stopModeBars, MaxHoldBars );

// When more symbols signal on the same bar than there are free slots, prefer
// the one where our fill assumption is most defensible. This is a liquidity
// preference, not a performance preference, and it was chosen before the first
// backtest was run.
PositionScore = AverageTurnover;

// ------------------------------------------------------------ chart view
// Applying the same file to a chart pane draws the rule on the price, so the
// trades in the report can be checked against something you can see. The arrows
// mark SIGNAL bars; the fills happen one bar later.
_SECTION_BEGIN( "First system" );

Plot( Close,     "Close",                colorDefault, styleCandle );
Plot( TrendLine, "MA(" + MaPeriod + ")", colorBlue,    styleLine | styleThick );

PlotShapes( shapeUpArrow   * Buy,  colorGreen, 0, Low,  -20 );
PlotShapes( shapeDownArrow * Sell, colorRed,   0, High,  20 );

_SECTION_END();

Download first-system.afl148 lines

How it works

The two rule numbers come first, with the reason they were chosen. 200 because it is the most widely quoted length, which makes it the length least likely to have been picked because it flattered this data. 60 because it is roughly a quarter of a trading year. Neither is claimed to be a good number; both are claimed to be numbers chosen without looking.

The settings section puts every backtester option into the file. Initial equity, position count, no margin, zero interest on idle cash, and commission as a percentage of trade value. SetPositionSize( PositionPct, spsPercentOfEquity ) expresses the slot size as a percentage of portfolio equity, which is the documented meaning of that constant. Ten slots at 10% each means a fully invested portfolio when ten candidates are held, and the guide is explicit that a position size must be set at all if you want more than one position open at a time.

The execution section applies one bar of delay to every signal and moves the fill against us by the slippage figure on both sides.

The filters apply the liquidity floor and the participation cap to the entry only. An exit is never filtered: if a symbol dries up while we are holding it, we still want out. Both filters are applied before ExRem(), because removing redundant signals first and filtering afterwards can strand an exit with no matching entry.

The time exit uses the backtester’s N-bar stop, ApplyStop( stopTypeNBar, stopModeBars, 60 ), which counts bars from the trade entry inside the engine. That is more reliable here than counting on a signal array, because SetTradeDelays() shifts the four signal arrays and nothing else — an exit produced by ApplyStop() is not one of those arrays, so it is generated from the bar the trade really began on.

PositionScore carries the tie-break. The guide documents that when more entry signals appear than MaxOpenPositions or available funds allow, AmiBroker ranks candidates by the absolute value of this variable. Average turnover is always positive, so higher means more liquid means preferred.

The chart section exists so that the same file can be applied to a chart pane and checked by eye. The arrows mark signal bars; the fills happen one bar later.

Key functions

• SetPositionSize( size, method ) — with spsPercentOfEquity, size is read as a percentage of portfolio-level equity. The other documented methods are shares, a dollar value, a percentage of the currently open position (for scaling only), and no change.
• SetOption( field, value ) — sets an Analysis setting from the formula. field is a string, so a typo is silently ignored and the option simply does not take effect.
• ApplyStop( type, mode, amount, exitatstop, volatile, ReEntryDelay, ValidFrom, ValidTo ) — here called with three arguments, matching the documented N-bar example. Part 28 covers the rest of the signature.
• ExRem( ARRAY1, ARRAY2 ) — returns 1 on the first true value in ARRAY1, then 0 until ARRAY2 is true, even if ARRAY1 is true again in between.

4. Running the first backtest

A checklist, in order. Deviating from it is fine; not recording the deviation is not.

1. Fix the range before you run anything. Choose the start and end dates and write them in your notes now. Choosing them afterwards, when you have seen how the equity curve looks, is a decision made on the data.
2. Open the Formula Editor, paste the formula, save it under a name you will recognise, then Tools → Send to Analysis.
3. In the Analysis window set Apply to to Filter and choose your watch list. Set Range to From-To dates and enter the dates from step 1.
4. Press the Settings button and confirm Periodicity is Daily. Everything else the system needs is already set from the formula and will override the dialog, but it is worth looking once so that you know what the dialog contains.
5. Press Backtest. The trade list appears in the results pane.
6. Press Report for the full statistics. The Report button’s drop-down opens the Report Explorer, which keeps every historical report rather than only the last one.
7. Save the report and the formula together, under the date. You will want to come back to this exact run in Part 28 and again in Part 31.

✦Tip

If you want to watch the selection process — which candidate won a contested slot and why — switch the report mode to the detailed log with SetOption( “PortfolioReportMode”, 1 ) before running. It is verbose, and it is the fastest way to understand what ranking is doing. Take it back out afterwards.

5. Reading the result without excitement

The order in which you read a report determines what you conclude from it. Reading it in the order the eye naturally goes — biggest number first — reliably produces enthusiasm. Use this order instead, every time, and write each figure down before looking at the next.

First: how many trades. The report gives counts for all trades, long trades and short trades in three columns. If your run produced 34 trades, nothing that follows means much, whatever it says. Sample size does not appear in any single metric and it governs the reliability of all of them.

Second: Exposure %. Documented as market exposure computed bar by bar — the sum of the per-bar exposures divided by the number of bars, where a bar’s exposure is the value of open positions divided by portfolio equity. A system that was 12% invested on average is not comparable to one that was 90% invested, and this figure is also the denominator of two other metrics, so a small value inflates them dramatically.

Third: Max. system % drawdown. The largest peak-to-valley percentage decline in portfolio equity, reported as a negative number. Read it as a question about yourself: could you have kept trading through that, for as long as it lasted? Part 34 returns to this, and Part 29 to the duration, which the headline figure does not show.

Fourth: Annual Return %. The compounded annual return, computed with proper compounding over calendar days between the first and last bar. Note the label: the report row is Annual Return %, not CAR and not CAGR, although the docs use CAR as the abbreviation in prose and inside CAR/MaxDD.

Fifth: CAR/MaxDD. Annual Return % divided by Max. system % drawdown. This is the number to prefer over the return alone, because it prices the return in units of the pain required to collect it. The docs describe values above 2 as good; treat that as a convention rather than a threshold, especially on a small trade sample.

Sixth: the trade statistics. Profit Factor (profit of winners divided by loss of losers), Payoff Ratio (average win over average loss), Avg. Profit/Loss and Avg. Profit/Loss % (which the docs also call expectancy in dollars and in percent), and Avg. Bars Held. Check Avg. Bars Held against your 60-bar cap: if it is close to 60, the time exit is doing most of the work and the reverse cross is nearly irrelevant, which is a finding about your rules.

Seventh: the equity curve, and specifically its flat periods rather than its slope.

!Important

None of these numbers is an expectation of future returns. They are a description of what these rules would have done, on this data, in this period, under these assumptions, if every simulated fill had been achievable. Every one of those qualifiers is load-bearing. A result you cannot state with all of them attached is a result you should not state at all.

The sentence you are allowed to write

Draft your conclusion as one sentence with every qualifier in it. Something of this shape:

On [watch list], over [start] to [end], with a 50-bar turnover floor of 5,000,000, trading at the next open with 0.10% slippage and 0.10% commission per side, 10 equally weighted slots and no stop, these rules produced [n] trades with an Avg. Profit/Loss % of [x], Exposure % of [y] and Max. system % drawdown of [z].

If you cannot fill in every bracket from your own notes, the missing one is a thing you did not record and should go back for. If the sentence feels deflating compared with the number you saw in the report, that is the correct emotional response and it is worth noticing.

6. Verifying it by hand

A report is only evidence if the trades in it are the trades your rules describe. Three checks, about ten minutes:

1. Pick three trades at random from the trade list. For each, open the symbol’s chart with the same formula applied. The entry date should be exactly one bar after a green arrow, and the entry price should be that bar’s open plus 0.10%.
2. Find a trade that lasted exactly 60 bars. There should be some, and their exits should be on the clock rather than at a cross. If there are none at all, the N-bar stop is not in force — check the ApplyStop() line.
3. Check a rejected symbol. Find a thinly traded symbol in your watch list that produced no trades, and confirm on its chart that its price did cross the average. If it did and no trade appears, the liquidity floor is working. If it did and a trade does appear, it is not.

=Expected result

A run over a broad watch list and ten or more years should produce a few hundred trades, an Exposure % well below 100 (the system is out of the market whenever fewer than ten symbols are above their averages), an Avg. Bars Held somewhere below the 60-bar cap, and a Max. system % drawdown large enough to be uncomfortable. The exact figures depend entirely on your universe, period and data, which is why this course does not print any.

7. Common errors

Symptom	Likely cause
Only one position is ever open	No position size set, so 100% of funds go into a single security. The guide states this explicitly
Many signals, very few trades	Slots full, or cash exhausted. In the default backtest mode, once a trade is skipped, later entries in that block are ignored until an exit appears
Entry prices are not the next bar’s open	SetTradeDelays() missing, or BuyPrice never assigned so the Settings trade price is in force
Commission appears to have no effect	Option name misspelled. SetOption() field names are strings and typos fail silently
Trades in symbols with tiny volume	Liquidity floor applied to the exit as well as the entry, or applied after ExRem()
No trades at all	Watch list not selected in Apply to; or the liquidity floor is above every symbol’s turnover; or the range predates enough history for a 200-bar average
Every trade lasts one bar	The entry was written as a state rather than an event, so Buy and Sell overlap
Results change between runs with no code change	A setting is coming from the dialog rather than the formula. Move it into the formula

8. Extensions

Each of these is a separate experiment with its own write-up, not an improvement to this one. Record each as a new dated entry in your log, and keep this run as the baseline.

• Sensitivity, not optimisation. Run the same system at 150 and at 250 bars, and at 40 and 80 bars of hold. You are not looking for the best of the six; you are looking at whether the result is a plateau or a spike. A spike is a warning; a plateau is mild reassurance.
• Cost doubling. Set slippage and commission to 0.20% each and re-run. Note which conclusions survive.
• Universe split. Run separately on the most liquid half and the least liquid half of the watch list. If the effect lives only in the thin half, your fill assumptions are carrying the result.
• Regime split. Run on two or three non-overlapping sub-periods. One good period and two bad ones is a very different finding from three mediocre ones.
• The benchmark. Compute what simply holding the universe over the same period would have done. A rule that returns less than that, with more work and more risk, has told you something clear.

9. Why we are not optimising yet

The system in front of you has two numbers in it. You could try twenty values of each, run 400 backtests in a few minutes, and take the best. Every instinct says to do it. Here is what would happen if you did.

The best of many is biased upward, by an amount you cannot see. Results vary across parameter values partly because of a real effect and partly because of noise. Taking the maximum selects for both, and the selected maximum overstates what that parameter set will do on new data. The bias grows with the number of combinations tried, and nothing in the report tells you how large it is — the report of the best run looks exactly like the report of a genuinely good system. Part 31 measures this properly.

You would have no baseline left. With an optimised result and nothing to compare it to, you cannot answer the only question that matters: how much of this came from the idea and how much from the searching? An unoptimised run of the same rules, on the same universe, over the same period, answers it directly. That run only exists if you do it first.

It teaches the wrong goal. An hour spent making a number larger trains you to treat the number as the objective. The objective is a defensible belief about how a rule behaves. Those two goals point in opposite directions surprisingly often: the honest response to a weak result is usually to abandon the idea, and optimisation always offers an alternative.

The parameters are not where the leverage is anyway. Position sizing, costs, the universe and the exit structure move results far more than the length of a moving average. Part 34 demonstrates this by running one signal set through three sizing models. Tuning the average first is optimising the smallest term.

And you would spend your out-of-sample data. Every additional variation you test on the same history consumes some of its capacity to surprise you. Data you have looked at forty times can no longer tell you that you were wrong. Part 32 builds the walk-forward machinery that manages this budget; until then, the cheapest way to preserve it is not to spend it.

?Reality check

“Optimising” is often presented as the serious, professional part of system development — the stage where an amateur idea becomes a real strategy. Held up to the light, that framing is backwards. Optimisation is a measurement technique for how sensitive a result is to its parameters, and it is a poor method for choosing them. Parts 31 to 33 teach it as the former. The single most useful thing you can do today is finish this project with the numbers you started with.

Recap and deliverables

You should now have six artefacts, all dated and stored together:

1. The hypothesis, written before anything was run, with its prior and its falsifier.
2. The rules table, which passed a solo ambiguity test.
3. The assumptions block, with the not-modelled list filled in.
4. first-system.afl, unmodified.
5. The backtest report, saved, with the range and watch list recorded.
6. Your one-sentence conclusion with every qualifier attached, plus a short list of the things the run does not settle.

That sixth item is the deliverable that distinguishes this from a hobby. A system you can describe honestly, including its limits, is one you can improve deliberately. A system attached to a number you cannot qualify is one you will keep adjusting until it stops being possible to tell whether it ever worked.

Part 28 opens the backtester properly: position sizing methods, cost modelling, stops, and the move from one symbol to a portfolio competing for one pool of capital. Part 29 reads the report in full. Part 30 is where this system gets attacked, and where you find out how much of it survives.

Check your understanding

Question 1. Why does the project use ApplyStop( stopTypeNBar, stopModeBars, 60 ) rather than counting bars on the Buy array?

• Because BarsSince() does not exist in AmiBroker
• Because SetTradeDelays() shifts the four signal arrays and nothing else, so a count taken from the trade entry inside the engine matches the trade that actually happened
• Because the N-bar stop is faster to compute
• Because ApplyStop() is required whenever position sizing is used

Show the answer and why

Answer: Because SetTradeDelays() shifts the four signal arrays and nothing else, so a count taken from the trade entry inside the engine matches the trade that actually happened

Delays shift Buy, Sell, Short and Cover after your formula has run. A bar count computed from the unshifted Buy array starts one bar before the trade did. The N-bar stop is generated by the backtester from the entry, so the two cannot disagree.

Question 2. A backtest report shows an Annual Return % of 21 and an Exposure % of 9. What should you notice first?

• That the system is efficient with capital and therefore better
• That with 9% average exposure, the metrics that divide by exposure will look extraordinary and mean very little on their own
• That the exposure figure must be a bug
• That the system needs more positions

Show the answer and why

Answer: That with 9% average exposure, the metrics that divide by exposure will look extraordinary and mean very little on their own

Exposure % is the denominator of both Risk Adjusted Return % and Net Risk Adjusted Return %, so a system that is nearly always in cash produces enormous values for both. Low exposure is neither good nor bad by itself — it changes what the other numbers mean, and it usually means very few trades, which is the thing to check next.

Question 3. Which of these are legitimate reasons for not optimising the two parameters now? Select all that apply.

• The best result out of many trials overstates what that parameter set will do on new data
• Without an unoptimised run you have nothing to compare an optimised one against
• Optimisation is never useful in system development
• Every variation tested on the same history spends some of its ability to surprise you

Show the answer and why

Answer: The best result out of many trials overstates what that parameter set will do on new data, Without an unoptimised run you have nothing to compare an optimised one against, Every variation tested on the same history spends some of its ability to surprise you

Optimisation is a legitimate technique — for measuring how sensitive a result is to its parameters, which is what Parts 31 to 33 teach. What it does poorly is choose them, and doing it first destroys the baseline that would let you see how much of the improvement was search rather than signal.

Question 4. The report shows Avg. Bars Held of 57 against a 60-bar time exit. What does that tell you about the rules?

• The time exit is broken
• The reverse cross is doing very little work; almost all trades are being closed by the clock
• The system is holding too long and should be shortened
• Nothing — average bars held is unrelated to the exit

Show the answer and why

Answer: The reverse cross is doing very little work; almost all trades are being closed by the clock

If the average holding period sits just under the cap, most trades are reaching it, so the exit you thought you were testing is barely firing. That is a finding about the rules rather than a defect: the system you ran is closer to "hold for 60 bars after a cross" than to "hold until the trend reverses", and any conclusion should say so.

Check my answersClear