* Wait for first session bar before filling equity market orders at open
EquityFillModel.MarketFill could fill a market order placed right after
market open using data from the previous trading date, because the first
bar of the current session has not been emitted yet. ShouldWaitForFreshData
only covered hour/daily resolutions, so minute/second orders filled on stale
prices.
Add IsWithinFirstResolutionSpanAfterMarketOpen: when the order time is within
the lowest subscribed resolution span after the open and the price is stale,
wait for the first bar instead of filling on the previous date's price.
* Share opening-bar stale-fill wait across fill models
Move IsWithinFirstResolutionSpanAfterMarketOpen to the base FillModel and add
a ShouldWaitForFreshDataOnStale sibling helper that combines it with the
existing coarse-resolution ShouldWaitForFreshData check. The base FillModel,
FutureFillModel and EquityFillModel market fills now share this single wait
decision at their stale-data guards.
ShouldWaitForFreshData is intentionally left untouched at its GetMarketFillPrice
call site, which uses it to choose the bar open vs current price and is not
gated by staleness, so fill prices for finer resolutions are unchanged. The
opening-bar helper is guarded against always-open markets, which have no
session open to wait for.
* Add regression algorithm for stale fill at market open
Reproduces the opening-bar stale fill issue: a market order placed one second
after the open while subscribed to minute resolution. Without the fix the order
fills on the previous trading date's stale price; the algorithm asserts in
OnOrderEvent that a fill never happens within the first minute after the open,
so it errors without the fix and passes with it.
Uses SPY minute data over 2013-10-07 to 2013-10-11, which is available in the
repository Data folder.
* Add unit tests for stale fill wait at market open
Cover the opening-bar stale fill scenario directly at the fill model level:
a market order placed within the first bar after the session open, while only
the previous session's stale bar is available, must wait instead of filling on
the stale price, and fills once the first session bar arrives. EquityFillModel
also asserts the boundary (orders past the first bar still fill on stale data),
and FutureFillModel covers the shared base helper from the future path.
* Generalize stale market-order fill wait to any time of day
Replace the market-open-specific wait with a generic check: a market order
that would be filled on stale data waits for fresh data when the latest
available data is more than one subscribed resolution bar behind the current
time. This no longer considers the market open explicitly; it covers the
opening bar (the first session bar has not been emitted yet) and any intraday
data gap larger than the resolution.
ShouldWaitForFreshDataOnStale now takes the latest data end time and the
current time instead of the order time, and is shared by FillModel,
FutureFillModel and EquityFillModel. Coarse resolutions (hour/daily) still
always wait; tick never waits. Internal configurations are included when
sizing the resolution bar. EquityFillModel's best-effort price helpers now
report the stale data end time so the gap can be measured.
Tests: EquityFillModelTests and FutureFillModelTests cover the market-open and
mid-session stale cases (wait then fill on fresh data) plus the within-one-bar
boundary (fill on stale). The regression algorithm is generalized to assert no
fill happens on data staler than the resolution, with orders at the open and
mid-session. Pre-existing plumbing/data-selection tests that used degenerate
timestamps were given fresh timestamps so they still exercise their original
intent.
* Add sample data and adjust regression algorithms for stale-fill wait
Add minute/daily sample data so market orders that now wait for fresh data
can fill (ES futures gap days, TWX/GOOG equities and options, SPXW weeklies,
GC futures/options copy for 2020-01-06). Adjust a few regression algorithms
to the deferred-fill behavior: cap orders in the extended-market continuous
future test, ignore daily-resolution SPY in the automatic-seed data checks,
and refresh OptionAssignmentStatistics expected constants.
* Update regression expected statistics for stale-fill wait
Regenerate ExpectedStatistics, DataPoints and AlgorithmHistoryDataPoints for
the regression algorithms affected by the wait-for-fresh-data fill change and
the added sample data: futures/options fill-timing shifts, ES data-point count
increases, and GOOG 2015-12-28 outcome changes.
* Trim SPXW sample data to expiries within filter window
The two SPXW algorithms filter with Expiration(0,7), so contracts expiring
more than a week out are never subscribed. Drop those far-dated expiries from
the 2021-01-06/08 minute files (760KB->108KB and 776KB->108KB on the quote
files). Fills, DataPoints and statistics are unchanged; both regression tests
still pass.
* Trim ES minute and GOOG option sample data to order-fill minimum
The ES minute gap-day files source no order fills (daily-resolution algos fill
from es_daily); keep only the front contract used for execution and drop the
unused back-month contracts. Trim the GOOG 2015-12-28 option file (no fill
depends on it) to the morning chain window. Regenerate the back-month futures
statistics affected by the dropped back-month bars. Full CSharp regression
suite passes (722/722).
* Use SMA gap threshold in BasicTemplateContinuousFuture for C#/Python parity
At a fast/slow SMA cross the two averages can coincide to within rounding
noise, where the C# (decimal) and Python (double) comparisons disagree,
producing different orders between languages. Require a minimum gap before
acting on a cross so both languages stay in lockstep, and update the shared
expected statistics accordingly.
* Mirror order cap in Python algorithm and update future history counts
Apply the same pre-2013-11-12/3-order cap to the Python
BasicTemplateContinuousFutureWithExtendedMarket algorithm for C#/Python parity,
and update the QuantBook future-history expected counts to reflect the added ES
sample data.
* Use SMA gap threshold in BasicTemplateContinuousFutureWithExtendedMarket for C#/Python parity
This algorithm had the same fast/slow SMA cross divergence already fixed in
BasicTemplateContinuousFutureAlgorithm (ad8fc33): at the 2013-10-29 cross the two
averages coincide to within rounding noise (C# decimal diff -1e-25, Python double
diff exactly 0.0), so the raw `_fast > _slow` / `_fast < _slow` comparisons disagree
between languages. C# fired a liquidate+rebuild that Python skipped, producing 5
orders in C# vs 3 in Python. Require a minimum 0.001 gap before acting on a cross so
both languages stay in lockstep, and regenerate the shared expected statistics
(Total Orders 5 -> 3).
* Document SMA cross threshold as a C#/Python parity workaround
Add a short note before the fast/slow SMA comparisons in both continuous-future
template algorithms clarifying that the minimum-gap threshold exists only so the
C# and Python versions take the exact same trades on the limited sample data in
the repository, where decimal vs double rounding can disagree at a cross.
* Fetch subscription configs once per equity market fill
MarketFill resolved the subscription configs twice per fill: once via the
best-effort price helpers (GetSubscribedTypes) and again via
ShouldWaitForFreshDataOnStale. Fetch them once and thread them through both
paths via optional parameters, leaving existing callers unchanged.
* Measure stale-fill wait against order submission time
ShouldWaitForFreshDataOnStale compared the latest data end time against the
security current time. Compare against the order submission time instead so the
decision to wait for fresh data reflects how stale the data is relative to when
the order was placed. Realign the stale-price warning fill test accordingly.
* Fix stale market data in SendingNewOrderFromOnOrderEvent test
The market price tick was timestamped a day before the order submission time,
so under the order-time staleness check the market orders waited for fresh data
instead of filling. Use a reference time with the tick one minute before the
order so the data is fresh and the orders fill.
* Centralize internal-inclusive subscription config lookup in fill models
ShouldWaitForFreshDataOnStale re-resolved the subscription configs through
the ShouldWaitForFreshData call it makes first, and GetMarketFillPrice did
the same. Thread the already-fetched configs through ShouldWaitForFreshData
and GetMarketFillPrice so each market fill resolves them at most once.
Add a GetSubscriptionDataConfigs(Security) helper on the base FillModel that
returns the internal-inclusive configs, and route every fill-model call site
through it to remove the duplicated lookup and repeated comment.
* Avoid list allocation in ShouldWaitForFreshData
Replace the Where(...).ToList() + All(...) with a single foreach over the
subscription configs, short-circuiting on the first non-coarse resolution.
Build & Test Lean / build (push) Has been cancelled
Regression Tests / build (push) Has been cancelled
Report Generator Tests / build (push) Has been cancelled
Research Regression Tests / build (push) Has been cancelled
Syntax Tests / build (push) Has been cancelled
Python Virtual Environments / build (push) Has been cancelled
* Implement a prototype of the maximum recovery time function.
* Add unit test skeletons.
* Add failing test
* Issue #4581: Implement MaxDrawdownRecoveryTime.
* Issue 4581: Add DTO for Drawdown Percentage, Drawdown Enddate, and High Value
* Issue 4581: Fix bgu for when lDrawdowns list is empty.
* Issue 4581: Change names of tests. Change name of file.
* Issue 4581: Make adjustements to flow of adding drawdowns to lDrawdowns.
* Issue 4581: Add multiple unit tests.
* Issue #4581: Change name of unit test
* Issue #4581: Add to PerformanceMetrics
* Issue #4581: Add Maximum Drawdown Recovery to PortolioStatistics class.
* Issue #4581: Add to portolfio statistics class.
* Issue #4581: Add to statistics builder.
* Issue #4581: Add report key.
* Case #4581: Convert to decimal.
* Issue #4581: Correct comment.
* Issue #4581: Correct performance metrics view model string.
* Case #4581: Correct statistics builder view model string..again.
* Issue #4581: Placed DradownDradownDateHighValueDTO at the end of the file for simpler diff.
* Issue #4581: Add 2 new tests.
* Issue #4581: Change algorithm so that when multiple maximum drawdowns occur, the longest of all recoveries is reported.
* Issue #4581: Add unit test.
* Issue #4581: Remove reportkey. Change dto name.
* Issue #4581: Change summary.
* Issue #4581: Change comment.
* Add max drawdown recovery calculation with unit tests
* Update regression algorithms with the new metric
* Solve review comments
* Update regression algorithms
* Add TryGet to safely get the key: MaximumDrawdownRecovery
* Ignore MaximumDrawdownRecovery metric in OptimizationBacktest Json
* Revert changes in Messaging
* Update regression algorithms
* Add test case: TakesLongestRecoveryAmongMultipleDrawdowns
* Use integer days for MaximumDrawdownRecovery
* Add MaximumDrawdownRecoveryReportElement
* Use more explicit names
* Rename files and variables for consistency
* Update regression algorithms
---------
Co-authored-by: Alain Schaerer <aschaerer@pcatg.com>
* Fix CA1819 and CA1002 warnings
Changed the type of Languages statistic in regression tests from
Language[] to List<Language>. By doing that, the warning CA1819 was
removed but then the warning CA1002 was raised. However, this warning
was expected to be excluded from QuantConnect.Algorithm.CSharp.
* Improve implementation
* Simplify code
* Fix bugs
* First draft of the solution
* Add missing changes
* Remove the new KPI's from report
* Fix bugs
* nit change
* Add improvements
* Fix regression tests
* Solve bugs in the regression algos
* Fix regression tests bugs
* Expand unit tests and add minor changes
Regression Tests / build (push) Has been cancelled
Benchmarks / build (push) Has been cancelled
Build & Test Lean / build (push) Has been cancelled
Research Regression Tests / build (push) Has been cancelled
Python Virtual Environments / build (push) Has been cancelled
* Add Sortino ratio to statistics and report
* Adds Sortino Ratio to Report Key Statistics
* Addresses Peer-Review
Reuse `SharpeRatioReportElement` and change the template.
* Reuse Calculations Across Statistics and PortfolioStatistics
* Adds Sortino Ratio to Regression Algorithms
* Removes Sortino Ratio from Optimization Result Table
---------
Co-authored-by: Alexandre Catarino <AlexCatarino@users.noreply.github.com>
Build & Test Lean / build (push) Has been cancelled
Regression Tests / build (push) Has been cancelled
Research Regression Tests / build (push) Has been cancelled
Python Virtual Environments / build (push) Has been cancelled
* Add new Order.PriceAdjustmentMode property
* Minor fix and unit test
* Minor fix and regression algorithms' stats update
* Unit test fixes
* Minor fix
* Set order price adjustment mode to raw always for live trading
Regression Tests / build (push) Has been cancelled
Python Virtual Environments / build (push) Has been cancelled
Benchmarks / build (push) Has been cancelled
Build & Test Lean / build (push) Has been cancelled
Research Regression Tests / build (push) Has been cancelled
* Refactor alpha statistics
- Refactor alpha statistics, cleaning up and simplifying no longer required calculations and scoring
- Adding new InsightEvaluator abstraction, adding C# & PY regression
algorithms
* Optimization backtest result json converter update
* Address reviews
- Remove IAlphaHandler, move insight storage responsability to IResultHandler
and centralizing insight collection on the QCAlgorithm.Insights to be
reused by the framework models
- Fix portfolio turnover single day backtests and duplicate time
sampling handling. Updating regression algorithms
* Add InsightCollection tests and minor fixes
* Adding more & improved tests
* add data count properties
* 'add history count property
* assert data counts
* update missing override
* consider override/virtual cases
* implement data count
* add message handler for regression tests
* use regression test message handler
* set algorithm manager for regression test message handler
* update data count
* check if stats are present, check if algo manager is not null
* update
* add c# algo
* make same as c# algo
* use new line
* logic shifted to RegressionTestMessageHandler
* cleanup
* auto cleanup
* skip non deterministic data count
* change data count
* use inheritance
* improve stats
* update couht
* add sma indicator to c# and customSMA to python
* call base method before executing further
* skip test
* revert to original
* add duplicate sma
* skip regression test
Regression Tests / build (push) Has been cancelled
Build & Test Lean / build (push) Has been cancelled
* Implement scheduled event sampling solution
* Use UTC time, only update daily portfolio value once a day
* For daily resolutions sample chart always
* Cleanup
* Drop resample daily all together
* Force final sample
* Regression updates
* FIx LiveResultHandler to update portfolio and benchmark values outside of sampling event
* Name the daily sampling event
* Address review pt 1
* Drop force and use reference wrapper
* Adjust tests
* Fix warning for Benchmark Timezone Misalignment and also add test
* Fix for daily resolution orders and test adjustments
* Also warn on universe settings with daily resolution
* Update missed regression
* Fix reference wrapper use
* Update regression after rebase
* Add values back in for Daylight Algo
* Have statistics builder skip day 1 performance
* Regression adjustments
* Test adjustments
* Update regression unit test
* Adjust some regressions starts to show performance values
* Add hourly algorithm for beta comparison
* Address missing Python regression changes
* Remove null comment
Regression Tests / build (push) Has been cancelled
Build & Test Lean / build (push) Has been cancelled
* Updates Equity Market Data
* Updates Unit Tests
* Updates Regression Tests
In this commit we include regression tests with small changes (slightly different CAGR, Alpha, etc, but same number of trades) due to the data update.
* Updates Regression Tests 2
The following regression tests were adapt because of verification of hard-coded market data values:
- `AdjustedVolumeRegressionAlgorithm`
- `HistoryWithSymbolChangesRegressionAlgorithm`
- `OptionRenameRegressionAlgorithm`
- `RawDataRegressionAlgorithm`
- `SwitchDataModeRegressionAlgorithm`
The following regression tests have more trades since adjusted prices allowed more 1-2 shares trades that were rounded down to zero before:
- `AddUniverseSelectionModelCoarseAlgorithm` 23 -> 35
- `MeanVarianceOptimizationFrameworkAlgorithm` 12 -> 14
- `PortfolioRebalanceOnDateRulesRegressionAlgorithm` 298 -> 324
- `PortfolioRebalanceOnInsightChangesRegressionAlgorithm` 83 -> 86
- `ScheduledUniverseSelectionModelRegressionAlgorithm` 86 -> 90
- `SectorExposureRiskFrameworkAlgorithm` 17 -> 22
- `SetHoldingsMultipleTargetsRegressionAlgorithm` 8 -> 9
- `StandardDeviationExecutionModelRegressionAlgorithm` 196 -> 199
- `UniverseUnchangedRegressionAlgorithm` 11 -> 17
- `VolumeWeightedAveragePriceExecutionModelRegressionAlgorithm` 237 -> 238
Especial cases:
- `BlackLittermanPortfolioOptimizationFrameworkAlgorithm` 18 -> 17
- BLM model sensibility
- `OptionChainedAndUniverseSelectionRegressionAlgorithm`
The following regression tests have different Capacity because of different volume from lowest capacity asset, except:
- `OptionEquityCoveredCallRegressionAlgorithm` New lowest capacity asset is underlying
- `OptionEquityCoveredPutRegressionAlgorithm` New lowest capacity asset is underlying
* Revert File Update for SPWR and SPWRA
* Fix Regression Tests
Temporarily removes python regression test for `MeanVarianceOptimizationFrameworkAlgorithm` as the `MeanVarianceOptimizationPortfolioConstructionModel` for each version are yeilding different results. If we use C# version in `MeanVarianceOptimizationPortfolioConstructionModel.py`, the results match.
* Changes Optimization Method in MinimumVariancePortfolioOptimizer [Py]
Uses `trust-constr` method.
See https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.minimize.html
Build & Test Lean / build (push) Has been cancelled
* Add property for capacity. Remove unused variable
* Move SymbolCapacity and CapacityEstimate to common, passed through Symbol to runtime statistics
* Add null checks
* Remove uninvested and untradable assets from capacty calculations
* Add SymbolCapacity influential period
* Updates Regression Tests
- DelistingEventsAlgorithm
- Allows additional contributions from delisted AAA.1
- DelistingFutureOptionRegressionAlgorithm
- Removes DC01H12 contributions one month later
- FutureOptionBuySellCallIntradayRegressionAlgorithm
- Allows additional contributions from future after expiry replacing the contribution of the next contract option
- DelistedFutureLiquidateRegressionAlgorithm
- FutureOptionCallITMExpiryRegressionAlgorithm
- FutureOptionCallITMGreeksExpiryRegressionAlgorithm
- FutureOptionPutITMExpiryRegressionAlgorithm
- FutureOptionShortCallITMExpiryRegressionAlgorithm
- FutureOptionShortPutITMExpiryRegressionAlgorithm
- FuturesAndFuturesOptionsExpiryTimeAndLiquidationRegressionAlgorithm
- Allows additional contributions from future after expiry
- FutureOptionCallOTMExpiryRegressionAlgorithm
- FutureOptionPutOTMExpiryRegressionAlgorithm
- FutureOptionShortPutOTMExpiryRegressionAlgorithm
- IndexOptionCallITMGreeksExpiryRegressionAlgorithm
- IndexOptionCallOTMExpiryRegressionAlgorithm
- IndexOptionShortCallOTMExpiryRegressionAlgorithm
- Allows additional contributions from option after expiry
- MACDTrendAlgorithm
- Removes contribution when SPY is not invested for over one month
- UniverseSelectionRegressionAlgorithm
- Allows additional contributions from delisted GOOAV replacing GOOG (new symbols)
* Adds Lowest Capacity Asset to Regression Tests
* Normalize expected value -0, because -0 is also written to file if updated
* Write Symbol.Value for lowestCapacitySymbol or empty string for empty Symbol
* Update Regressions
* Update 'Lowest Capacity Asset' to Symbol.ID
Co-authored-by: Jared Broad <jaredbroad@gmail.com>
Co-authored-by: Martin-Molinero <martin@quantconnect.com>
Co-authored-by: Colton Sellers <Colton.R.Sellers@gmail.com>
* Fixes Double to Decimal Cast in GetAnnualPerformance
`GetAnnualPerformance` raises an exception if the `AnnualPerformance` calculation returns a double that cannot be cast to decimal (smaller than `decimal.MinValue` or bigger than `decimal.MaxValue`).
See `ProbabilisticSharpeRatio` where the same solution was applied.
* Updates SPY Market Data
SPY is a key asset since it is the default benchmark, and any change can lead to different `Alpha` and `Beta`
* Updates Unit Tests to Reflect Data Update
* Updates Regression Tests to Reflect Data Update I
Most of the regression tests change because of updated data (market and factors) of SPY (default benchmark) while the total trade remain the same.
* Updates Regression Tests to Reflect Data Update II
The following regression tests were changed to adapt to adjusted prices and keep the total trades:
- `BacktestingBrokerageRegressionAlgorithm`
- `LimitIfTouchedRegressionAlgorithm`
- `PortfolioRebalanceOnCustomFuncRegressionAlgorithm`
- `SetAccountCurrencySecurityMarginModelRegressionAlgorithm`
- `StopLossOnOrderEventRegressionAlgorithm`
- `TimeInForceAlgorithm`
The following regression tests have more trades since adjusted prices allowed more 1-2 shares trades that were rounded down to zero before:
- `FreePortfolioValueRegressionAlgorithm` 2 -> 3
- `PortfolioRebalanceOnDateRulesRegressionAlgorithm` 291 -> 298
- `TrailingStopRiskFrameworkAlgorithm` 5 -> 7
Especial cases:
- `AutoRegressiveIntegratedMovingAverageRegressionAlgorithm` 65 -> 52
- ARIMA model sensibility
- `BlackLittermanPortfolioOptimizationFrameworkAlgorithm` 18 -> 19
- BLM model sensibility
- `ExtendedMarketHoursHistoryRegressionAlgorithm` 20 -> 18
- Less minute bars before market opens
* Addresses Peer-Review
Fix `BacktestingBrokerageRegressionAlgorithm` to use `CalculateOrderQuantity` and round down `quantity` to an even number to pass a value assertion and update the expected value from 50 to 52.
The quantity calculated by `CalculateOrderQuantity` has changed from 50 to 53 because of factor file update.
* Adds CapacityEstimate and SymbolCapacity
The capacity estimation has been moved from
the report generator and wired directly into
Lean via the ResultHandler. In addition,
the capacity estimation strategy has changed
to account for errors in the previous iteration
of the capacity estimation.
Many many thanks to Jared for being much of the
mastermind behind this project. It would have
been harder to complete without him to bounce ideas
off of.
* Moves old tests to regression algorithms
* Adds Estimated Capacity statistic
* Removes old capacity estimation tests
Final report capacity estimation. Pushing to save state
* Fixes bugs, cleans up code and adds comments
* Adds forced sampling to Capacity Estimation
* Misc. bug fixes for daily data
* Updates capacity test cases' Estimated Strategy Capacity statistic
* Adds Capacity Estimate to all regression algorithms
* Removes Report's StrategyCapacity class and fixes bug in tests
* Adds null check in BacktestingResultHandler to fix
BacktestingTransactionHandler failing tests
* Deletes old capacity estimation classes
* Retrieve capacity estimates from backtest statistics results
instead of calculating at runtime
* Make $0.00 capacity return as "-" and Result = 0 in report
* Adds capacity to runtime statistics
* Converts capacity to number denoted by financial figures in RuntimeStats
* Addresses review: code cleanup for Capacity and adds comments to regression tests
* Update OrderListHash to use MD5 as hash instead of hash code
* Update regression algorithm OrderListHash statistic
* Use full MD5 hash as OrderListHash, update regression statistic
* Fixes failing regression tests
The stale price message should use `ToStringInvariant` with `price.EndTime` for string represenation across different cultures. The former behavior has impact on the order list hash calculation.
- Reduce MinimumVariancePortfolioOptimizar precision goal so that both
CSharp and Py MeanVarianceOptimizationFrameworkAlgorithm return the same
results
- Limit factor file dates in factor file generator unit test
- EventProvider will receive start date during initialization, this will
be used by the `MappingEventProvider` to correctly set current mapped
symbol
- Adding unit test, updating existing regression test
- Replacing `BaseData.AdjustResolution` for `DefaultResolution` and
`SupportedResolutions`
- Making `Resolution` nullable for `Algorithm.AddData` methods
- The `ISubscriptionDataConfigService` will set the default resolution
if none was provided and assert it is supported
- Fix bug with `PythonData` `IsSparseData` and `RequiresMapping`
resolution
- Adding `BaseData.AdjustResolution()` that should return a valid
resolution for the given data and security type.
This allows us to set a limitation which is useful to avoid invalid data
requests or unnecessary fill forward situations. The user will be
notified through a console message.
- Adding unit and regression test
- Updating example algorithms custom data resolution
- Some performance improvements. Wont change console color if
`SelectedOptimization` is defined
* Deleted regression algorithms because they tested behavior similar to
other existing regression algorithms
* Fixed new bug in regression algorithm due to AddData changes
* Added unit tests for wrapt version and package existence
* Fix issue where data would be set to raw normalization mode
- Custom data types will know whether or not Lean should use map files
- Updating regression test with sample custom data using map files,
which can run locally
- Adding unit tests for the `SubscriptionDataReaderHistoryProvider`,
checking it mappes equities and options correctly
Jhonathan Abreu