程序代写代做代考 DTS 2018

DTS 2018

Dynamic trading
strategies/systems

Strategy types, position sizing, and
more

JEN-WEN LIN, PHD, CFA

Types of strategies

Mark S. Rzepczynski (1999), “Market Vision and Investment Styles:
Convergent versus Divergent Trading”, The Journal of Alternative
Investments, pp. 77-82.

Four core decisions for a
trading system

1. When to enter a position.
2. How large a position to take on.
3. How to get out of positions.
4. How much risk to allocate to different

sectors and markets.
Greyserman and Kaminski (2014), “Trend Following with Managed
Futures: The Search for Crisis Alpha”, John Wiley & Sons, Inc.

Position Sizing
1. Trading systems allocate capital across

asset classes/markets by systematically
allocating risk or capital to individual markets.

2. Position sizing must consider the volatility
of a particular market.

Remark: In your assignment, we don’t consider position sizing
but apply unity exposure, i.e. either long or short one unit of
asset.

1. Position sizing based on dollar risk1:
The nominal position (𝑣) is equal to a sizing function times
the total adjusted dollar risk times the nominal value of
one contract.

𝑣 = 𝑠 ×
𝜃 × 𝑐

𝜎((Δ𝑃) × 𝑃𝑉.///0///1
23245 4789:2;7 73554< <=:>

× (𝑃𝑉 × 𝑃)

1 Alternatively, we could consider average trading range (ATR) for each individual market. Average trading ranges are a simple
way to incorporate actual trading volatility and volumes as opposed to using realized volatility. See Clenow (2013) for more
details.

1. The sizing function (𝑠) is a number between –1 and 1
(𝑠 ∈ [−1,1]). It determines the size and direction of a
contract based on trading signals and trend strength.

2. The total adjusted dollar risk allocated is equal to the
allocated dollar risk divided by the futures contract dollar
risk. The allocated dollar risk is simply the risk loading (𝜃)
times the allocated capital (𝑐) per market. The futures
contract dollar risk is the realized dollar risk (𝜎((Δ𝑃) of
each contract price over a lookback window of time (𝐾)
times the point value (𝑃𝑉)).

2. Position sizing based on volatility target:
Guilleminot et al. (2014) suggests the trend-based
allocation for asset 𝑖 at time 𝑡 as

𝑤IJ = 𝐾I
𝑀𝑎𝑥N0, TrendIJ U

𝜎I
J ,

where 𝑇𝑟𝑒𝑛𝑑IJ denotes the trend signal for asset 𝑖 at time 𝑡,
𝜎IJ is the estimate of volatility for asset 𝑖 at time 𝑡, and the
parameter 𝐾I is calculated by

𝐾I =
Target volatility

𝜎I
,

where 𝜎I is the estimate of portfolio volatility at time 𝑡 with

the weight for asset 𝑖 equal to
cdeNf,g<;h7i jU ki j . Copyright © Jen-Wen Lin 2017 Page 11 What’s wrong with this design? Copyright © Jen-Wen Lin 2017 Page 12 John Moody et al. (1998), “Performance Functions and Reinforcement Learning For Trading Systems and Portfolios”, Journal of Forecasting, Vol. 17, pp. 441-470. Copyright © Jen-Wen Lin 2017 Page 13 Copyright © Jen-Wen Lin 2017 Page 14 Copyright © Jen-Wen Lin 2017 Page 15 Research project of past graduate student Dynamic Asset Allocation for Pairs Trading https://www.cs.toronto.edu/~francohtlin/dynamic- asset-allocation.pdf Copyright © Jen-Wen Lin 2017 Page 16 Unfinished discussion 1. Reinforcement learning/Dynamic programming 2. Signal generation (Data-driven/ value-based) 3. Long-run return predictability 4. Else (frequency, cost, risk management)

Related Posts