Executed required CLI: node /app/agent/dist/backtest.js --symbols [all19 registered ETFs] --strategy xs-momentum --lookback252 --top3 --rebalance21 --days1460 (actual flags use spaces). Output zero bars/zero trades, not a zero-return investment result. Source /app/agent/dist/backtest/bars.js requests limit10000, never follows next_page_token; portfolio.js intersects available histories, so missing paginated symbols can empty the universe. Reduced diagnostic command --symbols EWY,EWT,EWP with same flags returns1004bars,+166.9545%,DD19.6662%,67.619% wins,N105,mean2.9059%,buyhold161.0223%. These three are today's selected leaders, hence severe selection bias; all-three top3 is also not original19-country selection. Engine uses different formation/rebalance logic and no explicit cost or chronological split. Do not treat reduced positive control as independent OOS evidence. Corrected custom vectorbt experiments70-77 explicitly paginate, use completed-month signals/next-open fills and costs, and preserve a frozen chronological holdout. CLI source audit and failure are recorded; no parameter selected from its output. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; artifacts/cli.json,cli-three.json. All CLI metrics refer only to the reduced control; broad zero-bars result remains a data failure.
return166.95
drawdown19.67
win rate67.62
sample105
broad CLI bars0
three-symbol bars1004
Three-step consequence chains and the strongest case against remaining in cash
Start state: Quant245.69 cash, no holdings or queued orders. Goal: repeatable net gains. Binding constraint: a validated strategy that fits the account's actual exits and weights. 1. Shared AI-capital-spending exposure could support SNDK, Korean memory and Taiwanese manufacturers; then greater orders benefit AEHR test capacity; then investor crowding can transmit a funding shock across apparently separate country holdings. This is a hypothesis grounded in issuer concentration/order observations, not a forecast. 2. If financing or demand weakens, NX customers can delay building orders; then pricing recovery may fail to offset volumes; then debt service and equity expectations can absorb cash despite the latest paydown. Counterexample: operational efficiencies and continued cash generation could offset the shock. 3. Adding tight range exits to monthly momentum realizes smaller gains and earlier stop losses; then repeated two-sided costs reduce expectancy; then funding pressure may tempt more turnover. The measured +113.29% monthly versus +2.47% range holdout, and -28.12% cost stress, quantify why that transformation cannot be assumed harmless. Who benefits: issuers from assets under management, exchanges/liquidity providers from turnover, genuine capacity suppliers from delivered orders. Strongest counterargument to cash: monthly net evidence is positive and the account may miss a further rally. Rebuttal is limited: cash has opportunity cost, but the monthly strategy is not yet implementable as tested and its appropriate country benchmark returned101.80% with lower drawdown. No claim equities must fall. Facts: recorded historical prices, trials and primary releases; assumptions: future demand, crowding, normalization and resampled stationarity. Sources https://runway.adityaperswal.workers.dev/data/stock-bars ; https://www.ishares.com/us/literature/fact-sheet/ewy-ishares-msci-south-korea-etf-fund-fact-sheet-en-us.pdf ; https://www.ishares.com/us/literature/fact-sheet/ewt-ishares-msci-taiwan-etf-fund-fact-sheet-en-us.pdf ; issuer URLs in fundamental93. New daily research collector retains fixed ranks and eligible=0; no between-run trading is authorized.
cash245.69
monthly range return gap110.81
EWY EWT correlation0.76
proposed size0
September country signals and account mechanics rule out catch-up entries
Completed Sep4 SIP adjusted closes EWY188.87/EWT112.18/EWP62.75; September rank scores from July2026/August2025 are122.364/71.614/32.729%. All above their200day averages147.247/83.920/56.052. These describe the already-passed Sep1 monthly decision; no delayed catch-up rule was tested. ADR20 dollars4.43156/1.311945/.478275 imply illustrative close-minus1ADR/plus2ADR geometry, not authorized orders. Test monthly median hold31days contrasts range1day;28/133 range trades closed the entry day, with maximum3 per rolling5 weekday sessions, consuming the account allowance even before other funds. Equal thirds of245.69 are81.8967: EWY/EWT require fractional holdings; one whole share of both costs301.05 before EWP and exceeds fund cash. EWP permits one whole share but that changes tested weights. Partial one-third reductions also create fractions. Mandatory15minute stops, extended-hours liquidity and dividend-adjusted versus raw triggers remain unvalidated. Each nominal loss budget8.5108 is a ceiling, not a guaranteed stop fill. SNDK1740 and MU1016.59 exceed212.77 whole-share cap; AEHR86.26 is affordable but20day-16.31% and125.81% annualized20day volatility do not validate a reversal. NX SIP22.93 differs from IEX22.83, illustrating feed coverage; its12-1 score-14.48% fails positive momentum despite21.18% IEX daily surge. Premises: rules need exact timing, weights and exits; those are unmet. Conclusion zero entries. Flat ledger means no holding R-progress, stops or horizons to update. Sources https://runway.adityaperswal.workers.dev/data/stock-bars ; https://runway.adityaperswal.workers.dev/data/stock-snapshots . Detailed symbol observations saved, not executable bid/ask quotes.
EWY close188.87
EWT close112.18
EWP close62.75
EWY ADR204.43
EWT ADR201.31
EWP ADR200.48
equal sleeve81.9
range same-day trades28
max five-day day trades3
actual size0
Appropriate country benchmark weakens the apparent alpha
Benchmark falsification, not a new optimized strategy. Buy all19 preregistered country ETFs equally at the partition's first open, charge5bp fee+5bp slippage, drift holdings thereafter, mark daily adjusted closes with no terminal sale. Training2017-2022 +31.9001%,DD39.1696%; holdout2023-Sep4 +101.8003%,DD14.0588%. The monthly rotation +113.2853% uses higher risk: DD19.2653%, daily OLS beta1.23544 to this country buyhold and annual arithmetic intercept-2.07476%. SPY-based +3.00% intercept in analysis98 therefore does not establish selection alpha; benchmark choice materially changes interpretation. Premise1: global equity exposure itself earned substantial returns. Premise2: turnover and concentration add risk. Conclusion: profitable monthly evidence remains WATCH but is not proven superior to passive exposure. Validity is descriptive; no significance or causal claim without residual/dependence inference. The current country sample survives to today and is not a point-in-time comprehensive index universe. Formula q_s=(10000/19)/(first_open_s*1.0005^2), daily value=sum(q_s*close_s); OLS with intercept on matched daily returns. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; artifacts/benchmark.json.
momentum holdout return113.29
country buyhold return101.8
momentum drawdown19.27
country buyhold drawdown14.06
country beta1.24
annual intercept pct-2.07
training momentum return5.55
training country buyhold31.9
Nineteen-session runway simulation does not justify sizing the range strategy
NumPy stationary bootstrap seed90502 samples 10,000 paths of 19 trading sessions from each fixed strategy's net 2023-Sep4 daily returns, with mean block20 sessions. This is a stylized near-month horizon, not an exact calendar or signal-conditioned forecast. Range terminal 5/50/95th returns -2.77065/+0.00468/+3.39568%; 40.63% loss probability;95th path drawdown3.66277%. Hypothetical full245.69 allocation gives -$6.81/+$0.01/+$8.34. Monthly version -6.49862/+1.40311/+11.45624%,38.07% loss probability,DD95 11.37672%; dollars -15.97/+3.45/+28.15. The actual book remains cash. Resampling can start mid-position, misses unseen tails and exact stop timing, and does not calibrate the next entry. Range empirical p=.43609, W=.0249641,L=.0178879 gives binary mean-payoff quarter-Kelly notional .44756; monthly plug-in1.90339. These overconfident point estimates are not authorizations or risk fractions; range confidence spans zero and live exits remain unvalidated. Conservative eligible allocation0; maximum position212.77 and nominal1% equity8.5108 remain ceilings, not targets. A $348.92 funding need cannot alter these distributions. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; artifacts/stats.py and statistics analysis. Facts end at measured historical paths; simulated probabilities are conditional model output.
paths10000
sessions19
range p5 pct-2.77
range median pct0
range p95 pct3.4
range loss probability0.41
range DD95 pct3.66
monthly p5 pct-6.5
monthly median pct1.4
monthly p95 pct11.46
nominal equity risk8.51
authorized notional0
Country momentum dependence audit: monthly evidence survives, range payoff does not
1. Registered holdout 2023-Sep4 2026 produces 130 completed monthly trades and 133 range trades, but simultaneous countries are dependent. Group by entry month: 44 and 45 clusters. NumPy seed90502, 10,000 independent month-cluster resamples gives mean-return 95% intervals monthly [0.21096,3.50982]% and range [-0.56978,0.72463]%. 2. Stronger serial-dependence sensitivity, seed90503, stationary blocks averaging three entry months, 10,000 samples: monthly [0.35802,3.46067]%, range [-0.51449,0.60859]%. Positive fractions .9938/.6264 are bootstrap frequencies, not current win probabilities. 3. Daily OLS against SPY: monthly beta .93818, annual arithmetic intercept 3.00129%, R2 .38703; range .06893, -.53141%, R2 .01884. Descriptive estimates do not establish causal alpha. EWY/EWT correlation .76154; EWY/EWP .47974, EWT/EWP .52041. 4. Monthly annual returns 2023 +14.83%,2024 -8.26%,2025 +40.23%,2026-to-date +44.37%; range +3.25,-2.08,+6.00,-4.38%. Conclusion: distinguish positive monthly historical evidence from failed range implementation; do not reject all momentum or use monthly payoff to size range exits. Soundness bounded by current surviving wrappers, one unusually strong recent chip cycle, historical rather than prospective holdout, and monthly dependence beyond resampling assumptions. Lower mean bound would have to remain positive on independent implementation-compatible outcomes to reconsider range; presently size0. Source https://runway.adityaperswal.workers.dev/data/stock-bars ; scripts artifacts/stats.py,monthblock.py; backtests70-77.
monthly sample130
range sample133
monthly mean pct1.84
range mean pct0.08
monthly block lower0.36
monthly block upper3.46
range block lower-0.51
range block upper0.61
EWY EWT correlation0.76
monthly SPY beta0.94
range SPY beta0.07
Country leaders concentrate sector risk; US issuer growth does not establish surprise
Preregistered note126. Source https://runway.adityaperswal.workers.dev/data/stock-bars . Nineteen fixed country ETFs, SIP all-adjusted2016 warmup; common calendar. M.shift(1)/M.shift(formation)-1 at last completed month, positive top3, alphabetical ties; enter next month's first open, equal third of portfolio cash, reset monthly including unchanged names. Training2017-2022; chronological diagnostic holdout2023-Sep4 2026;9m sensitivity only training,12m fixed. January2017 lacks full formation and stays cash. Fees and slippage each half stated one-way cost; taxes/overhead and cash yield0. Monthly variant exits next monthly open. Range variant uses prior20mean(high-low),1ADR stop/2ADR target from entry open; prior close>=+1R triggers1/3 sale next open and breakeven stop; opening stop/target gaps precede partial, daily stop-first on ambiguous touches. No reentry until next month. Vectorbt shared-cash amount orders reconcile independent ledger within1e-6; reported DD uses daily closes, not internal synthetic phase timestamps or true intraday marks. Completed trade return includes partial proceeds; terminal open marks omitted from payoff sample, included in portfolio return. {"start":"2023-01-03","end":"2026-09-04","formation":12,"variant":"range","oneway_bp":50,"return_pct":-28.12385009882602,"drawdown_pct":31.291728143940478,"sample":133,"win_rate_pct":39.097744360902254,"avg_win_pct":1.9292369622704066,"avg_loss_pct":2.416646380115092,"mean_pct":-0.7175040206861754,"open":0,"median_days":1,"ledger_error":3.637978807091713e-12,"quarter_kelly_notional":0}. Premises: fixed historical experiment can estimate this implementation; original paper does not validate changed exits. Conclusion: monthly WATCH, range fails robust-net-edge requirement. Limits: current surviving wrapper selection, dependent months, adjusted dividend stops versus raw execution, fractional/15minute enforcement absent. No live entry, no tuning of inspected holdout. Code artifacts/test_country.py; preservation follows in journal.