blvresearch

research tools to be used with Bluevalor database

######### PORTFOLIO #########

General description: provides basic objects for data backtesting:

• StockReturns
• PortfolioDates
• PortfolioStrategy
• Portfolio
• PortfolioCharacteristics

Portfolio formation procedure: STEP 1. write a "strategy" class, that inherits from PortfolioStrategy by overwriting _get_starting_points method with an algorithm of your choice

    one can use any type of data available in Bluevalor Model data output

STEP 2. Initiate the object from STEP 1 providing Bluevalor Model data output

STEP 3. Initiate Portfolio class providing result of STEP 2 (i.e. your strategy) on init. Now you can operate with your portfolio by e.g. checking portfolio members over time, or calculating portfolio performance.

STEP 4. Initiate PortfolioCharacteristics class by providing the result of STEP 3 on init. This will allow you to check different statistics of your portfolio (check description of PortfolioCharacteristics class)

PortfolioStrategy after initiating gives access to the following: .output = Bluevalor Model data output .positions = time series with lists of entities for each rebalancing day any global attributes defined in the class that inherits from PortfolioStrategy .HOLDING_PERIODS = how many periods should entities be held in portfolio .PAUSE_PERIODS = how many periods should pass between determining portfolio positions and starting to actually hold them .REBALANCING_FREQUENCY = frequency of rebalancing periods .PORTFOLIO_SIZE = number of entities held in a portfolios

Explanation of RankingPeriod/RebalancingDay/PausePeriods/HoldingPeriods relationship:

Rebalancing day vs holding period: rebalancing day is the day when we decide about the construction of our portfolio for the next holding period, we should hold this portfolio starting at the earliest on the next day this is obtained by shifting portfolio.members by 1 position

################ PortfolioDates: ################

• determines days when security selection is performed

• determines days when portfolio construction is changed

• takes a number of variables on init: 0. date_index = set of dates that are only taken into account when when determining rebalancing days and ranking days

  1. ranking_periods = number of periods used to calculate certain characteristics, to determine which securities to include in portfolio in the nearest future
  2. pause_periods = number of periods we should wait after determining the next portfolio structure, before actually holding it
  3. holding_periods = number of periods we will be holding a given set of stocks in our portfolio
  4. date_resampling_frequency = frequency with which initial date index should be resampled

• if pause_periods = 0 => ranking_days will overlap with rebalancing_days, since ranking_days are BACKWARD looking (at the end of a given day X we look back at certain characteristics of that day and any earlier days), while rebalancing_days are FORWARD looking (at the same end of a given day X, after deciding about portfolio structure, we rebalance our portfolio with action being taken before the start of the next trading day, so that already the next trading day return is based on that portfolio we constructed one day before; to be more realistic, we should specify a certain number of pause_periods)

• what is the last ranking day? it should be such a date close to the end of full date index, so that there is still place (a necessary number of days) to calculate/accrue interest on this last portfolio; last ranking day cannot be the last day of the total date index, and it shouldn't also be the day after which there's only place for pause periods; therefore last day is the day which is

• setting RANKING_PERIODS = 0: i) should it be possible? temporarly impossible

• resampling the date index adds days (with NaN values) for equal periods, this shouldn't have substantial impact on portfolio performance if any, since we will only extend the last period if it doesn't end at the end of last resampled period; this will extend last resampled period and leave empty cells

CASES: (used in testing as well)

1. daily rankings of securities, daily rebalancings of portfolios, no pause periods, bottom line case; on day 02.01.2013 after exchange close rank securities, based on a certain characteristic (e.g. return) over that day; after determining this ranking, construct portfolio and aquire it for the next trading day, i.e. hold this portfolio over 03.01.2013; at the end of 03.01.2013 repeat ranking, and construct new portfolio to be aquired immediately and held over 04.01.2013, and so on and so forth; last ranking perform on the day before last day (17.01.2013) and aquire this portfolio to hold it over 18.01.2013

   To summarize: we rank stocks on each trading day's evening (from 02.01.2013 to 17.01.2013) and immediately afterwards we order this portfolio; so that we hold it over the next trading day (assumptions!!); ranking days are exactly the same as rebalancing days

2. daily security ranking, daily rebalancing, one day pause between ranking and rebalancing; slightly different case compared to (1) use first day data to rank securities, without pause we would construct portfolio with such securities immediately after determining the ranks and buy/sell it to hold over the next day; in this case we apply a 1 day pause period, hence rebalancing will occur at the end of next day (X + 1) with day X + 2 in mind for this first portfolio; using X + 1 day data (the one when we had pause, and our portfolio was empty) we perform the 2nd ranking, which will be used for day X + 3 (since day X + 2 will be a pause for this second portfolio mixture)

   To summarize: we rank stocks on each trading day's evening (from 02.01.2013 until 16.01.2013); the differences compared to (1), we apply pause period after each ranking evening, hence we rebalance our portfolio on the next day's evening after ranking day, to hold it over the next day: 02.01.2013 - rank stocks after market close, decide on your portfolio construction 03.01.2013 - pause period, we do not hold this portfolio here yet, after market closes on that day we order the portfolio on which we decided one evening earlier; 04.01.2013 - we hold over this day the portfolio on which we decided about on 02.01.2013 (evening)

   The last ranking day is 16.01.2013 (evening), when we decide about the portfolio we will order next evening (17.01.2013) to hold over 18.01.2013 The lasst rebalancing day is 17.01.2013; this portfolio's return is measured over 18.01.2013

3. ranking based on 5 periods (last 5 trading days), one trading day pause and a given portfolio is held over the next 3 days before the next rebalancing; in this example we use 02.01.2013, 03.01.2013, 04.01.2013, 07.01.2013 and 08.01.2013 to perform comparison of securities and rank them (pay attention: 05.01.2013 and 06.01.2013 are omitted since no trading happened on these days); after exchange close on 08.01.2013 we know which stocks to hold in our portfolio but before constructing it we wait one trading day (09.01.2013) and we order this portfolio on 09.01.2013 (evening) to hold it over next, three trading days (10.01.2013, 11.01.2013, 14.01.2013); here again, weekend days are omitted; in the meantime second ranking took place, namely, on 11.01.2013 evening we considered last 5 trading days and created the ranking, the next trading day (14.01.2013) is used as a pause for this portfolio, and therefore we rebalance our portfolio after market close on 14.01.2013 and hold it over 15.01.2013, 16.01.2013, 17.01.2013; for this short period last ranking is performed after market close on 16.01.2013, taking into consideration last 5 trading days (10.01.2013, 11.01.2013, 14.01.2013, 15.01.2013, 16.01.2013); the next trading day (17.01.2013) is used as a pause for this portfolio, we order this portfolio on 17.01.2013 evening and hold it over the next day (one day only), 18.01.2013; because in this toy example we don't have more data

   To summarize: rankings were performed on 08.01.2013, 11.01.2013 and 16.01.2013; for each of those portfolio plans we used one-day breaks: 09.01.2013, 14.01.2013, 17.01.2013, and we rebalanced our portfolio in the evenings of those PAUSE days, to hold it on the next trading day (+ 2 more days, since in this example we use 3 holding periods for each portfolio we construct)

4. this case includes weekly (weeks ending Fridays) rebalancing, which means the last period is extended if it ends on any other day, but Friday, in this case (02.01.2013 to 31.01.2013 period) it means we will have 5 periods: Period 1: ending 04.01.2013 Period 2: ending 11.01.2013 Period 3: ending 18.01.2013 Period 4: ending 25.01.2013 Period 5: ending 01.02.2013 We use 1 period for ranking, 1 period for pause and 1 period for holding; therefore first ranking day is 04.01.2013 (evening, after market close); we rank securities based on this one period data, over the next period (whole week ending next Friday, 11.01.2013) we do nothing, and only then, after market closes on 11.01.2013 (first rebalancing day) we order portfolio to hold over the next (3rd period); At the end of Period 2 we perform the 2nd ranking, we wait with this selection over Period 3, and only at the end of Period 3 we order this selection to hold it over Period 4; The last ranking is performed at the end of Period 3, using Period 3 data; We wait with this selection over Period 4 and only at the end of Period 4 we order this portfolio and hold it over Period 5; We need to pay attention to the fact, that Period 5, although ends on 01.02.2013, doesn't contain ful data, namely no data fo 01.02.2013 To summarize: we have three ranking days. 04.01.2013, 11.01.2013, 18.01.2013 we have three pause periods in the background; having pause periods does not mean we have some discontinuities in return generation; not at all, starting with the first portfolio, we hold some stocks all the time until the last period

5. case 5 uses the same periods as case 4, the difference here is that we use two periods to rank securities and we do not have pause period, i.e. we order (rebalance) the portfolio immediately after we decide what to order and we hold this portfolio over the very next period; first ranking day is 11.01.2013, since we use Period 1 and Period 2 to decide about our portfolio selection; first rebalancing day is 11.01.2013, when we order portfolio to hold over entire Period 3; at the end of Period 3 (18.01.2013) we use Periods 2 and 3 data to perform another ranking (second ranking day) and we rebalance our portfolio immediately (second rebalancing day), and hold this portfolio over Period 4; at the end of Period 4 we perform our last ranking (using Periods 3 and 4 data) and we rebalance our portfolio for the last time and hold it over last Period 5; To summarize: we have 3 ranking days (11.01.2013, 18.01.2013, 25.01.2013) and 3 rebalancing days (11.01.2013, 18.01.2013, 25.01.2013)

6. in this case we use monthly resampled data, we take our daily data from 02.01.2013 until 03.07.2013 and we resample it monthly to get 7 periods, this example illustrates the issue with expanding last period to its full length, even if there's only 1 data point available for that period (here we only have 3 data points for July 2013, but we construct full monthly period anyway; this won't have any substantial impact on results and is convenient, therefore we use this approach); another characteristic in this case is we use month-ends, no matter if this last day of a month is a trading day, or not Period 1: 31.01.2013 Period 2: 28.02.2013 Period 3: 31.03.2013 Period 4: 30.04.2013 Period 5: 31.05.2013 Period 6: 30.06.2013 Period 7: 31.07.2013 first ranking day is 31.01.2013, we use January data to select stocks; the next period (February) is a break, and on 28.02.2013 we order the first portfolio which we will hold over the next two periods; therefore first rebalancing day is 28.02.2013 and we hold this selection over Periods 3 and 4 (betweend 01.03.2013 and 30.04.2013); the next ranking is performed on 31.03.2013 using March data, this second selection of stocks will be ordered only one period later (second rebalancing day: 30.04.2013) and held over Periods 5 and 6 (between 01.05.2013 and 30.06.2013); the last ranking is performed on 31.05.2013 using May data, this selection waits over the whole Period 6 and is ordered on the last rebalancing day (30.06.2013) and held over Period 7 (only 3 days of July though) To Summarize: in this setting we had three ranking days (31.01.2013, 31.03.2013 and 31.05.2013) which were based on previous month data we had 3 rebalancing days which were one period (one month) later after each ranking day (28.02.2013, 30.04.2013, 30.06.2013)

7. this case is a special one, since it uses zero ranking periods; if we use zero ranking periods we also need to use zero pause periods, because it makes no sense otherwise; such a case is a toy case, where we use a certain security selection immediately and hold it over the very first period available; then we can rebalance this portfolio based on the length/amount of holding periods; if we use as input ranking_periods = 0, we won't be able to see ranking_days of our PortfolioStrategy, because there are none; we are only able to see rebalancing_days, for which the same logic applies as in cases 1 to 6; in example case 7 we have 7 periods after resampling our date index: Period 1: 02.01.2013 Period 2: 03.01.2013 Period 3: 04.01.2013 Period 4: 07.01.2013 Period 5: 08.01.2013 Period 6: 09.01.2013 Period 7: 10.01.2013 first rebalancing should happen before trading on 02.01.2013 starts, so that we hold first security selection over Periods 1 and 2; this means we need to add one data point at the very beginning, i.e. first rebalancing day will be 01.01.2013, since we have holding_periods = 1, we will rebalance our portfolio daily, so: 1st rebalancing: 01.01.2013 (hold over 02.01.2013); date point added 2nd rebalancing: 02.01.2013 (hold over 03.01.2013) 3rd rebalancing: 03.01.2013 (hold over 04.01.2013) 4th rebalancing: 04.01.2013 (hold over 07.01.2013) 5th rebalancing: 07.01.2013 (hold over 08.01.2013) 6th rebalancing: 08.01.2013 (hold over 09.01.2013) 7th rebalancing: 09.01.2013 (hold over 10.01.2013)

   • why having ranking_periods = 0 and pause_periods != 0 makes no sense? if we have e.g. six periods: Period 1: 31.01.2013 Period 2: 28.02.2013 Period 3: 31.03.2013 Period 4: 30.04.2013 Period 5: 31.05.2013 Period 6: 30.06.2013 having ranking_periods = 0 means we do not rank securities based on the data over a certaing period, this means we can immediately decide on our selection on the very first day and hold this selection over the very first period, e.g. by choosing 10 securities randomly; if we applied pause periods into this setting this would mean we wouldn't order this portfolio for the very first period, but wait over Period 1 and order this selection for Period 2; then either applying pause will disappear after Period 1, since usually after first pause period it was hidden, and we had continuity in holding securities, or we will apply pause after each holding period ends and before we rebalance, which would mean we would have discontinuities in holding securities

8. last case uses similar setting to (7), but holding_periods = 2; therefore again first rebalancing occurs before the first returns happen, on 01.01.2013, this selection will be held over 02.01.2013 and 03.01.2013; on 03.01.2013 we rebalance and hold this selection over 04.01 and 07.01; on 07.01.2013 we rebalance and hold this selection over 08.01 and 09.01; on 09.01.2013 we rebalance and hold this selection over 10.01.2013 only Thus we have 4 rebalancing days: 01.01.2013, 03.01.2013, 07.01.2013 and 09.01.2013

################ STATIC STRATEGY ################ StaticStrategy is a PortfolioStrategy where Portfolio changes only at specific points in time determined by the following strategy attributes:

• ranking periods = number of periods used to compare certaing characteristics of stocks
• pause periods = number of periods to wait after determining the portfolio for the next holding period before actually holding it
• holding periods = number of periods to hold a given portfolio before next rebalancing
• rebalancing frequency = frequency of periods used, influences the meaning of all the above three attributes

StaticStrategy based on the above attributes determines:

• ranking days: days when we perform analysis of stock attributes to check which stocks will form our portfolio at the next rebalancing day
• rebalancing days: days when portfolio is changed (by assumption we change portfolio structure on the evening of the day before the first trading day when this portfolio is going to be active)
• positions: most important, as for PortfolioStrategy

Portfolio size can also bounded by default.

############## STOCK RETURNS ##############

Stock returns are contained in a class StockReturns, which takes total output and optional return type (by default absolute return is used). Initialized StockReturns give access to .monthly, .weekly, and .daily returns.

######### BACKTEST #########

Any kind of backtest class should perform only "overview" type of operations using Portfolio/PortfolioStrategy objects

################# DYNAMIC STRATEGY ################# DynamicStrategy is a PortfolioStrategy where Portfolio can be adjusted at every point in time (depending on specified frequency) according to a specified methodology.

DynamicStrategy checks at every periods end for every stock if it matches specified requirements, and if so, it adds this stock to a portfolio (possible pause period before actually adding the stock can also be specified) and holds it for a specified amount of holding periods

########## PLOTTING: ##########

provides infrastructure to create plots of data coming from testing Bluevalor's data