I've come across what appears to be a bug in WLP 5.6.

It can be replicated by selecting the strategy "OscillatorPulback - SingleEntry" and any reasonable time frame. I'm using daily bars and anywhere from 2 to 10 years in length. For completeness, I'm using PosSizer "Max Entries per Day" with 2 entries allowed and 8% of equity.

When I look at the raw returns by month, and order the data by month, I get multiple entries in the latter months. For example, using 2 years worth of data on a dataset, I get the following returns on the period starting:
12/1/2009 - $735.30
12/3/2009 - ($275.20)
12/3/2009 - ($374.10)
12/3/2009 - $1587.30

Multiple months have multiple entries, but all do not.

Obviously with reports like this, all data is suspect. I have seen this behaviour on multiple datasets, strategies, and PosSizers.

Has this been reported yet?

QUOTE:
Obviously with reports like this, all data is suspect.

I would rather say, one out of the two dozen performance visualizers ;) But I would do it only if what you're seeing could be reproduced, and currently I'm unable to do so. Could you let us know what DataSet you're running the Strategy on? Maybe it's the key to reproduce the issue. tia.

One of the datasets that I observed this on was comprised of EOD data on the following equities:

AAPL ACN ADTN AEA ALB ALGT ALTR ALXN AMD AMGN ANR APH ARO ATK BBL BCR BDX BEN BHP BIDU BIIB BKE BLUD BMY BUCY BYI CAGC CBST CEG CELG CEPH CHKP CHRW COCO COH CPTS CRI CSC CSCO CTCM CTRP CTSH DAR DCI DECK DEER DGX DLB DOX DRQ DVA EBAY EDU EJ EL ELNK ESV EV EXPD EZPW FAST FCX FDS FMC FOSL FRX GES GOOG GPS GRMN GWW GYMB HANS HAS HEW HMA HSP HTLD IDCC INFA INFY IPXL ISRG JCOM JNJ JWN KCI LSTR LULU LXK LZ MA MAT MDT MED MLHR MMM MSFT MXB NFLX NIHD NKE NSM NSR NTAP NVO OII OLN OMN ORCL PAY PCL PCLN PDCO PETS PKG PLL PNRA PSSI QCOR QSII RIMM RINO RMD RSH RTP RYN SAP SAPE SCCO SEIC SNI STE STJ SY SYK TIF TRA TRLG TSL TSS TV TXN URBN VALE VAR VPRT VRSN WAT WDR WIN WIT WMS WOOF XLNX XRAY YZC

Thanks for looking into this, Eugene.

I wasn't able to reproduce this either on this DataSet or on others, either breaking down by Month or by Day, using either intraday data scaled to Daily or true Daily data.

I show a 12/1/2009 and a 12/4/2009. There must be something in the data affecting this... will take a closer look.

Regarding the data: I've downloaded the majority of the symbols today from scratch (90% weren't present in any DataSets). Also, I see no trades during December (actually for a year now.)

In my case, it appears this was caused by an out-of-sequence bar for AMD:

CODE:
Please log in to see this code.

This condition was fixed by reloading the chart history, and it fixed the duplicate months in the by Period returns. I'm not sure why this should have the effect it does on the By Period returns, since those are based on the equity curve, but the general answer is "GIGO".

Cone nailed the problem - in the data. After tediously reloading the chart history for each symbol, the discrepancy went away. Thanks for pointing me in that direction.

However, there must be a better way to reload the data for an entire dataset. One way would be to delete the dataset and then create another with the same symbols, but I'm not certain that would reload the data. What is the best way to force a reload on all symbols in a dataset?

Thanks.

QUOTE:
After tediously reloading the chart history for each symbol

Or you could simply run the bundled "Bad History Data Check" strategy (Utilities). It does check for things like dates out of order.

QUOTE:
One way would be to delete the dataset and then create another with the same symbols,

No, this will not do anything to the symbol data.

QUOTE:
However, there must be a better way to reload the data for an entire dataset.

If the symbols are not present in other DataSets, you can delete their names from the DataSet, check the option to delete symbols not present in DataSets (Data Manager), and perform "Update All Data for Selected Providers", making sure the respective provider is checked. This will delete the files for these symbols, and you can paste back their tickers in the DataSet window.

Alternatively, you could open the folder where the data are stored, and delete the *.WL files manually.

Thanks for the insight. "Bad History Data Check" certainly beats the pants off of a click-happy carpal tunnel fun fest.