스트래티지 (Strategy) 만들기
|Language:||English • 日本語 • 한국어 • Русский|
strategy는 (매수나 매도를 하기 위한) 주문을 실행, 변경, 취소할 수 있는 스터디입니다. 코딩된 알고리즘에 따라서, strategy에서는 백테스팅 (backtesting) (과거 데이터로 strategy 이뮬레이션) 및 포워드테스팅 (forwardtesting) (실시간 데이터로 strategy 이뮬레이션)을 실행할 수 있습니다.
Pine Script 언어로 쓰여진 strategy는 Pine 인디케이터와 같은 모든 기능을 갖추고 있습니다. strategy의 코드를 쓸 때는 “study” 대신 “strategy” 키워드로부터 시작해야 합니다. Strategies는 시각적인 플로팅 뿐만 아니라, 주문의 실행, 변경, 취소도 할 수 있습니다. 특정 키워드로 strategy 성능 정보에 액세스 할 수 있습니다. 같은 정보를 “스트래티지 테스터 (Strategy Tester)” 탭에서도 볼 수 있습니다. 과거 데이터로 strategy 계산한 결과는 가상 체결로 나타나게 됩니다.
간단한 스트래티지 보기
//@version=2 strategy("test") if n>4000 strategy.entry("buy", strategy.long, 10, when=strategy.position_size <= 0) strategy.entry("sell", strategy.short, 10, when=strategy.position_size > 0) plot(strategy.equity)
스크립트가 컴파일되어 차트에 적용되면 차트 위에 곧 바로 주문 체결 마크가 나타나게 되며 백테스팅을 하는 과정에서 귀하의 밸런스가 변하는 것을 볼 수 있게 됩니다 (에쿼티 커브). 위 스트래티지는 모든 바에 대해 바이/셀을 하는 가장 간단한 스트래티지입니다.
첫줄에 보이는 strategy("test")는 이 코드가 스트래티지라는 것과 그 이름이 “test”라는 것을 말해줍니다. strategy.entry()는 “바이” 및 “셀” 오더를 내는 명령입니다. plot(strategy.equity)는 에쿼티 커브를 플로팅해 줍니다.
차트에 스트래티지를 거는 법
strategy를 테스트하려면 차트에 strategy를 걸어야 합니다. 테스트하고자 하는 심볼과 타임 인터벌을 정하십시오. “인디케이터 & 스트래티지" 다이얼로그 박스에 나오는 빌트인 스트래티지를 골라 쓰거나 파인 에디터에서 귀하 스스로 만들어 쓸 수도 있습니다.
Note: When using non-standard (Renko, Kagi, Line Break, Point and Figure, Heikin Ashi, Spread Charts) types of chart as a basis for strategy, you need to realize that the result will be different. The orders will be executed at the prices of this chart (e.g.for Heikin Ashi it’ll take Heikin Ashi prices (the average ones) not the real market prices). Therefore we highly recommend you to use standard chart type for strategies.
Backtesting and forwardtesting
On TradingView strategies are calculated on all available historical data on the chart and automatically continue calculation when real-time data comes in.
Both during historical and real-time calculation, code is calculated on bar closes by default.
If this is forwardtesting, code calculates on every tick in real-time. To enable this, check off the option “Recalculate On Every Tick” in settings or do it in script itself: strategy(..., calc_on_every_tick=true, ...).
You can set the strategy to perform additional calculation after an order is filled. For this you need to check off “Recalculate After Order filled” in settings or do it in script itself: strategy(…, calc_on_order_fills=true, ...).
There is a broker emulator on TradingView for testing strategies. Unlike real trading, the emulator fills orders only at chart prices, that is why an order can be filled only on next tick in forwardtesting and on next bar in backtesting (or later) after strategy calculated.
As stated above, in backtesting strategy is calculated on bar’s close. The following logic is used to emulate order fills:
- If opening price of bar is closer to highest price of the same bar, the broker emulator assumes that intrabar price was moving this way: Open → High → Low → Close.
- If opening price of bar is closer to lowest price of the same bar, the broker emulator assumes that intrabar price was moving this way: Open → Low → High → Close.
- Broker emulator assumes that there were no gaps inside bar, meaning all intrabar prices are available for order execution.
- If the option “Recalculate On Every Tick” in strategy properties is enabled (or strategy(..., calc_on_every_tick=true, ...) is specified in script), code is still calculated only on bar’s close, following the above logic.
Here is the strategy demonstrating how orders are filled by the broker emulator:
strategy("History SAW demo", overlay=true, pyramiding=100, calc_on_order_fills=true) strategy.entry("LE", strategy.long)
This code is calculated once per bar by default, on its close, however there is an additional calculation as soon as an order is filled. That is why you can see 4 filled orders on every bar: 2 orders on open, 1 order on high and 1 order on low. This is backtesting. If it were in real-time, orders would be executed on every new tick.
It is also possible to emulate order queue. The setting is called “Verify Price For Limit Orders” and can be found in strategy properties or set in script itself: strategy(... backtest_fill_limits_assumption=X, ...). The specified value = number of points/pips (minimum price movements), default value = 0. A limit order is filled if current price is better (higher for sell orders, lower for buy orders) for the specified number of points/pips. The execution price still matches the limit order price.
backtest_fill_limits_assumption = 1. Minimum price movement = 0.25.
A buy limit order is placed at price 12.50.
Current price is 12.50.
The order cannot be filled at current price only because backtest_fill_limits_assumption = 1. To fill the order the price must become 0.25*1 lower. The order is put in queue.
Assume that the next tick comes at price 12.00. This price is 2 points lower, what means the condition “backtest_fill_limits_assumption = 1” is satisfied, so the order should be filled. The order is filled at 12.50 (original order price), even if the price is not available anymore.
Order placement commands
All keywords that are designed for strategies start with “strategy.” prefix. The following commands are used for placing orders: strategy.entry, strategy.order and strategy.exit:
- strategy.entry — this command places only entry orders. It is affected by pyramiding setting (in strategy properties) and by strategy.risk.allow_entry_in keyword. If there is an open market position when an opposite direction order is generated, the number of contracts/shares/lots/units will be increased by the number of currently open contracts (script equivalent: strategy.position_size + quantity). As the result, the size of market position to open will be equal to order size, specified in the command strategy.entry.
- strategy.order — this command places both entry and exit orders. It is not affected by pyramiding setting and by strategy.risk.allow_entry_in keyword. It allows you to create complex enter and exit order constructions when capabilities of the strategy.entry and strategy.exit are not enough.
- strategy.exit — this command allows you to exit a market position by an order or or form multiple exit order strategy (stop loss, profit target, trailing stop). All such orders are part of the same strategy.oca.reduce group. An exit order cannot be placed if there is no open market position or there is no active entry order (an exit order is bound to ID of an entry order). It is not possible to exit a position with a market order using the command strategy.exit. For this goal the following commands should be used: strategy.close or strategy.close_all. If number of contracts/shares/lots/units specified for the strategy.exit is less than the size of current open position, the exit will be partial. It is not possible to exit from the same entry order more than 1 time using the same exit order (ID), that allows you to create exit strategies with multiple levels. In case, when a market position was formed by multiple entry orders (pyramiding enabled), each exit orders is bound to each entry order individually.
//@version=2 strategy("revers demo") if n > 4000 strategy.entry("buy", strategy.long, 4, when=strategy.position_size <= 0) strategy.entry("sell", strategy.short, 6, when=strategy.position_size > 0) plot(strategy.equity)
The above strategy constantly reverses market position from +4 to -6, back and forth, what is shown by its plot.
strategy("exit once demo") strategy.entry("buy", strategy.long, 4, when=strategy.position_size <= 0) strategy.exit("bracket", "buy", 2, profit=10, stop=10)
This strategy demonstrates the case, when market position is never closed, because it uses exit order to close market position only partially and it cannot be used more than once. If you double the line for exiting, the strategy will close market position completely.
//@version=2 strategy("Partial exit demo") if n > 4000 strategy.entry("buy", strategy.long, 4, when=strategy.position_size <= 0) strategy.exit("bracket1", "buy", 2, profit=10, stop=10) strategy.exit("bracket2", "buy", profit=20, stop=20)
This code generates 2 levels of brackets (2 take profit orders and 2 stop loss orders). Both levels are activated at the same time: first level to exit 2 contracts and the second one to exit all the rest.
The first take profit and stop loss orders (level 1) are in one OCA group. The other orders (level 2) are in another OCA group. It means that as soon as an order from level 1 is filled, the orders from level 2 are not cancelled, they stay active.
Every command placing an order has ID (string value) — unique order identifier. If an order with same ID is already placed (but not yet filled), current command modifies the existing order. If modification is not possible (conversion from buy to sell), the old order is cancelled, the new order is placed. strategy.entry and strategy.order work with the same IDs (they can modify the same entry order). strategy.exit works with other order IDs (it is possible to have an entry order and an exit order with the same ID).
To cancel a specific order (by its ID) the command strategy.cancel(string id) should be used. To cancel all pending orders the command strategy.cancel_all() should be used. Strategy orders are placed as soon as their conditions are satisfied and command is called in code. Broker emulator doesn’t execute orders before next tick comes after the code was calculated, while in real trading with real broker, an order can be filled sooner. It means that if a market order is generated at close of current bar, it is filled at open oif next bar.
//@version=2 strategy("next bar open execution demo") if n > 4000 strategy.order("buy", strategy.long, when=strategy.position_size == 0) strategy.order("sell", strategy.short, when=strategy.position_size != 0)
If this code is applied to a chart, all orders are filled at open of every bar.
Conditions for order placement (when, pyramiding, strategy.risk) are checked when script is calculated. If all conditions are satisfied, the order is placed. If any condition is not satisfied, the order is not placed. It is important to cancel price orders (limit, stop and stop-limit orders).
Example (for MSFT 1D):
//@version=2 strategy("Priced Entry demo") c = year > 2014 ? nz(c) + 1 : 0 if c == 1 strategy.entry("LE1", strategy.long, 2, stop = high + 35 * syminfo.mintick) strategy.entry("LE2", strategy.long, 2, stop = high + 2 * syminfo.mintick)
Even though pyramiding is disabled, these both orders are filled in backtesting, because when they are generated there is no open long market position. Both orders are placed and when price satisfies order execution, they both get executed. It is recommended to to put the orders in 1 OCA group by means of strategy.oca.cancel. in this case only one order is filled and the other one is cancelled. Here is the modified code:
//@version=2 strategy("Priced Entry demo") c = year > 2014 ? nz(c) + 1 : 0 if c == 1 strategy.entry("LE1", strategy.long, 2, stop = high + 35 * syminfo.mintick, oca_type = strategy.oca.cancel, oca_name = "LE") strategy.entry("LE2", strategy.long, 2, stop = high + 2 * syminfo.mintick, oca_type = strategy.oca.cancel, oca_name = "LE")
If, for some reason, order placing conditions are not met when executing the command, the entry order will not be placed. For example, if pyramiding settings are set to 2, existing position already contains two entries and the strategy tries to place a third one, it will not be placed. Entry conditions are evaluated at the order generation stage and not at the execution stage. Therefore, if you submit two price type entries with pyramiding disabled, once one of them is executed the other will not be cancelled automatically. To avoid issues we recommend using OCA-Cancel groups for entries so when one entry order is filled the others are cancelled.
The same is true for price type exits - orders will be placed once their conditions are met (i.e. an entry order with the respective id is filled).
strategy("order place demo") counter = nz(counter) + 1 strategy.exit("bracket", "buy", profit=10, stop=10, when = counter == 1) strategy.entry("buy", strategy.long, when=counter > 2)
If you apply this example to a chart, you can see that the exit order has been filled despite the fact that it had been generated only once before the entry order to be closed was placed. However, the next entry was not closed before the end of the calculation as the exit command has already been triggered.
Closing market position
Despite it is possible to exit from a specific entry in code, when orders are shown in the List of Trades on StrategyTester tab, they all are linked according FIFO (first in, first out) rule. If an entry order ID is not specified for an exit order in code, the exit order closes the first entry order that opened market position. Let’s study the following example:
strategy("exit Demo", pyramiding=2, overlay=true) strategy.entry("Buy1", strategy.long, 5, when = strategy.position_size == 0 and year > 2014) strategy.entry("Buy2", strategy.long, 10, stop = strategy.position_avg_price + strategy.position_avg_price*0.1, when = strategy.position_size == 5) strategy.exit("bracket", loss=10, profit=10, when=strategy.position_size == 15)
The code given above places 2 orders sequentially: “Buy1” at market price and “Buy2” at 10% higher price (stop order). Exit order is placed only after entry orders have been filled. If you apply the code to a chart, you will see that each entry order is closed by exit order, though we did not specify entry order ID to close in this line:
strategy.exit("bracket", loss=10, profit=10, when=strategy.position_size == 15)
strategy("exit Demo", pyramiding=2, overlay=true) strategy.entry("Buy1", strategy.long, 5, when = strategy.position_size == 0) strategy.entry("Buy2", strategy.long, 10, stop = strategy.position_avg_price + strategy.position_avg_price*0.1, when = strategy.position_size == 5) strategy.close("Buy2",when=strategy.position_size == 15) strategy.exit("bracket", "Buy1", loss=10, profit=10, when=strategy.position_size == 15) plot(strategy.position_avg_price)
- It opens 5 contracts long position with the order “Buy1”.
- It extends the long position by purchasing 10 more contracts at 10% higher price with the order “Buy2”.
- The exit order (strategy.close) to sell 10 contracts (exit from “Buy2”) is filled.
If you take a look at the plot, you can see that average entry price = “Buy2” execution price and our strategy closed exactly this entry order, while on the TradeList tab we can see that it closed the first "Buy1" order and half of the second “Buy2”. It means that the no matter what entry order you specify for your strategy to close, the broker emulator will still close the the first one (according to FIFO rule). It works the same way when trading with through broker.
It is possible to put orders in 2 different OCA groups in Pine Script:
- strategy.oca.cancel - as soon as an order from group is filled (even partially) or cancelled, the other orders from the same group get cancelled. One should keep in mind that if order prices are the same or they are close, more than 1 order of the same group may be filled. This OCA group type is available only for entry orders because all exit orders are placed in strategy.oca.reduce.
//@version=2 strategy("oca_cancel demo") if year > 2014 and year < 2016 strategy.entry("LE", strategy.long, oca_type = strategy.oca.cancel, oca_name="Entry") strategy.entry("SE", strategy.short, oca_type = strategy.oca.cancel, oca_name="Entry")
You may think that this is a reverse strategy since pyramiding is not allowed, but in fact both order will get filled because they are market order, what means they are to be executed immediately at current price. The second order doesn’t get cancelled because both are filled almost at the same moment and the system doesn’t have time to process first order fill and cancel the second one before it gets executed. The same would happen if these were price orders with same or similar prices. Strategy places all orders (which are allowed according to market position, etc).
The strategy places all orders that do not contradict the rules (in our case market position is flat, therefore any entry order can be filled). At each tick calculation, firstly all orders with the satisfied conditions are executed and only then the orders from the group where an order was executed are cancelled.
- strategy.oca.reduce - this group type allows multiple orders within the group to be filled. As one of the orders within the group starts to be filled, the size of other orders is reduced by the filled contracts amount. It is very useful for the exit strategies. Once the price touches your take-profit order and it is being filled, the stop-loss is not cancelled but its amount is reduced by the filled contracts amount, thus protecting the rest of the open position.
- strategy.oca.none - the order is placed outside of the group (default value for the strategy.order and strategy.entry commands).
Every group has its own unique id (the same way as the orders have). If two groups have the same id, but different type, they will be considered different groups. Example:
//@version=2 strategy("My Script") if year > 2014 and year < 2016 strategy.entry("Buy", strategy.long, oca_name="My oca", oca_type=strategy.oca.reduce) strategy.exit("FromBy", "Buy", profit=100, loss=200, oca_name="My oca") strategy.entry("Sell", strategy.short, oca_name="My oca", oca_type=strategy.oca.cancel) strategy.order("Order", strategy.short, oca_name="My oca", oca_type=strategy.oca.none)
“Buy” and “Sell” will be placed in different groups as their type is different. “Order” will be outside of any group as its type is set to strategy.oca.none. Moreover, “Buy” will be placed in the exit group as exits are always placed in the strategy.oca.reduce_size type group.
It is not easy to create a universal profitable strategy. Usually, strategies are created for certain market patterns and can produce uncontrollable losses when applied to other data. Therefore stopping auto trading in time should things go bad is a serious issue. There is a special group of strategy commands to manage risks. They all start with the strategy.risk.* prefix.
You can combine any number of risks in any combination within one strategy. Every risk category command is calculated at every tick as well as at every order execution event regardless of the calc_on_order_fills strategy setting. There is no way to disable any risk rule in runtime from script. Regardless of where in the script the risk rule is located it will always be applied unless the line with the rule is deleted and the script is recompiled.
If on the next calculation any of the rules is triggered, no orders will be sent. Therefore if a strategy has several rules of the same type with different parameters, it will stop calculating when the rule with the most strict parameters is triggered. When a strategy is stopped all unexecuted orders are cancelled and then a market order is sent to close the position if it is not flat.
Furthermore, it is worth remembering that when using resolutions higher than 1 day, the whole bar is considered to be 1 day for the rules starting with prefix “strategy.risk.max_intraday_”
Example (MSFT 1):
//@version=2 strategy("multi risk demo", overlay=true, pyramiding=10, calc_on_order_fills = true) if year > 2014 strategy.entry("LE", strategy.long) strategy.risk.max_intraday_filled_orders(5) strategy.risk.max_intraday_filled_orders(2)
The position will be closed and trading will be stopped until the end of every trading session after two orders are executed within this session as the second rule is triggered earlier and is valid until the end of the trading session.
One should remember that the strategy.risk.allow_entry_in rule is applied to entries only so it will be possible to enter in a trade using the strategy.order command as this command is not an entry command per se. Moreover, when the strategy.risk.allow_entry_in rule is active, entries in a “prohibited trade” become exits instead of reverse trades.
Example (MSFT 1D):
//@version=2 strategy("allow_entry_in demo", overlay=true) if year > 2014 strategy.entry("LE", strategy.long, when=strategy.position_size <= 0) strategy.entry("SE", strategy.short, when=strategy.position_size > 0) strategy.risk.allow_entry_in(strategy.direction.long)
As short entries are prohibited by the risk rules, instead of reverse trades long exit trades will be made.
TradingView strategies can operate in the currency different from the instrument currency. NetProfit and OpenProfit are recalculated in the account currency. Account currency is set in the strategy properties - the Base Currency drop-down list or in the script via the strategy(..., currency=currency.*, ...) keyword. At the same time, performance report values are calculated in the selected currency.
Trade profit (open or closed) is calculated based on the profit in the instrument currency multiplied by the cross-rate on the Close of the trading day previous to the bar where the strategy is calculated.
Example: we trade EURUSD, D and have selected EUR as the strategy currency. Our strategy buys and exits the position using 1 point profitTarget or stopLoss.
//@version=2 strategy("Currency test", currency=currency.EUR) if year > 2014 strategy.entry("LE", true, 1000) strategy.exit("LX", "LE", profit=1, loss=1) profit = strategy.netprofit plot(abs((profit - profit)*100), "1 point profit", color=blue, linewidth=2) plot(1 / close, "prev usdeur", color=red)
After adding this strategy to the chart we can see that the plot lines are matching. This demonstrates that the rate to calculate the profit for every trade was based on the close of the previous day.
When trading on intra-day resolutions the cross-rate on the close of the trading day previous to the bar where the strategy is calculated will be used and it will not be changed during whole trading session.
When trading on resolutions higher than 1 day the cross-rate on the close of the trading day previous to the close of the bar where the strategy is calculated will be used. Let’s say we trade on a weekly chart, then the cross rate on Thursday’s session close will always be used to calculate the profits.
In real-time the yesterday’s session close rate is used.