EFTA02673877.pdf

From: Lesley Groff Sent: Friday, February 7, 2014 7:23 PM To: Jeffrey Epstein Subject: Fwd: ATorus - Daily Portfolio Report 2/6 Attachments: Atorus_BacktestNAV_020614.pdt Untitled attachment 00921.htm Sent from my iPhone Begin forwarded message: =b>From: Michael Fowler c » Date: February 7, 2014, 1:57:21 PM EST To= Lesley Groff < • Subject: ATorus - Daily Portfolio Report 2/6 A Reminder About Security Selection & Position Sizing=/i>" Having displayed the "vol day" adjusted returns yesterday, I f=el it worth reminding about security selection and position sizing. Specifi=ally, the large winners, are not driven by out-sized position sizing (at in=eption) or a bias to small or mid cap securities becoming large cap securit=es. I've previously outlined our liquidity and market capitalization requir=ments in our Trading Assumptions document. Our position sizing=at inception, yields equal potential profit irrespective of notional dollar= at risk. Stated another way we eliminate the volatility "basis" risk betwe=n any positions, so that the denominators are all indexed to the same poten=ial impact to NAV. We then add to winners and never to losers. At the end o= the day, our assumption (yes, it is an assumption) is that the distributio= of returns, IN VOL DAYS and over a given interval of time, follows a Paret=-Iike distribution. By "indexing" our position sizing (e.g. Kelly Criterion=like) to vol, we are always "in" the positions that represent the majority o= returns and scale those returns by adding to them, without dollar cost ave=aging into losers. In essence, would you think the results are more stable o= someone who made 50% in a year even with a high Sharpe, wherein the sample=size was (i) small in the number of positions and factors; (ii) profit fact=r driven by a small subset of the total trades, and (iii) driven by excess p=sition sizing; or someone who made 15% in a year, wherein the sample size w=s (i) large; (ii) profit factor driven by top 25% of positions that do not r=peat; and (iii) position sized equally? While the former is possible, the o=ds are in the negative in terms of future consistency. Someone will do it, b=t the ability to ascribe the results to randomness or intelligence will be d=fficult. "A Near Constant Distribution: Exponents of the D=Ita in One Period Realized Volatility at the Next Moment Conditional on the=Previous Moment" A foundation of the st=ategy is how the distribution of the exponents of volatility scaling condit=onal on itself (T+1 (absolute realized vol)/(absolute average realized vol)= is nearly constant across any interval of time or system . This insight, a=lows for a constraint on the range of outcomes at the next interval. Concur=ently, this alters the return EFTA_R1_01953027 EFTA02673877 profile (as vol changes) over varying interva=s of times (what I call a Vol & Time Basis Risk). Having a varia=le that is nearly stationary, even in the range of its outcomes, allows one=to manage a complex dynamic system more prudently. "It is a capital mistake to theorize before one has dat=."Michael J. Fowler Intl. M=bile Work Email - Tradin Desk Email <mailt• <=iy style="font-family:ariallont-size:small"> The information containe= in this electronic mail message is confidential information intended only f=r the use of the individual entity named above, and may be privileged. If t=e reader of this message is not the intended recipient, you are hereby noti=ied that any dissemination, distribution, or copying of this message is str=ctly prohibited. If you have received this communication in error, please i=mediately notify us by telephone, and delete the original message. =/html>= 2 EFTA_R1_01953028 EFTA02673878