Style Model
|
Mid Blend
|
||
Sector Model
|
XLU
|
2.42%
|
|
Large Portfolio
|
Date
|
Return
|
Days
|
ABX
|
4/11/2013
|
-12.96%
|
318
|
NEM
|
9/30/2013
|
-15.75%
|
146
|
EW
|
10/28/2013
|
-11.45%
|
118
|
JOY
|
11/18/2013
|
0.21%
|
97
|
SWM
|
12/31/2013
|
-9.03%
|
54
|
TM
|
2/3/2014
|
1.38%
|
20
|
RS
|
2/10/2014
|
0.73%
|
13
|
CSCO
|
2/12/2014
|
-2.55%
|
11
|
INT
|
2/18/2014
|
-0.38%
|
5
|
CBI
|
2/20/2014
|
-0.99%
|
3
|
(Since 5/31/2011)
|
|||
S&P
|
Annualized
|
12.05%
|
|
Sector Model
|
Annualized
|
26.14%
|
|
Large Portfolio
|
Annualized
|
27.46%
|
Rotation: selling SWM; buying BT.
Given the robust performance of the Sector Model compared to
the Full Model, it’s time to take a look at the Sector Model’s performance in
more detail.
First, here is the Year to Date performance:
And this is compared to a 15 year back-test:
That last chart represents the following return rates:
S&P
|
Sector
|
|
Total
|
52.57%
|
1560.49%
|
Annualized
|
2.82%
|
20.35%
|
For most folks this is good enough, and there has been some
interest off site over the past several years to use the Sector Model as a
standalone fund. I’ll have more detail
next week about that separate fund, but it’s enough for now to mention that the
Year to Date returns represent trades in that fund.
Tim
Hi Tim. It'll be interesting to see the real-life sector model performance vs a benchmark (SPY,IWM) for a longer period of time. I recall a relatively significant under-performance when XLU was in effect for a while in 2013.
ReplyDeleteThat was a useful underperformance, helping me create the current version of the model. I'll post current live performance at least once each month going forward.
ReplyDeleteHere are some ideas for both live and backtested performance tracking stats of the sector model. These stats would be calculated on the selected XL* sector segments in comparison with both the benchmark (SPY) as well as the other 8 non-selected XL* sectors for each segment.
ReplyDeleteModel vs SPY:
(1) Total return of model vs total return of SPY (with and without tax impact)
(2) Percentage of model segments beating SPY in respective segment times
(3) Longest sequence of model selections beating/trailing SPY performance
(4) same as #3 on a calendar days scale rather than enumerated sequence
Model vs XL*:
(1) Total return of model vs total return of a posteriori-selected best XL* (in each segment)
(2) Histogram of relative model selections rank (among 9 XL* choices) for all segments
(3) Longest sequence of model selections being above-average/below-average XL* choices
(4) Calculate the above stats with and without tax impact
I don't see what Model vs XL could tell us. Perfection isn't that useful for a benchmark.
ReplyDeleteAs for Model vs SPY, I've been looking at the following:
Months:
SPY Sector Advantage
Average% 0.47% 1.68% 1.21%
Median% 1.06% 1.31% 0.25%
Best% 10.92% 24.68% 13.76%
Worst% -16.52% -15.64% 0.88%
Years:
SPY Sector Advantage
Average% 6.45% 21.02% 14.57%
Median% 10.70% 21.85% 11.15%
Best% 32.31% 58.07% 25.76%
Worst% -36.80% -16.37% 20.43%
That's for the backtest. I can do the same for the live performance of the fund, but that's only 2 months old so far.
A bit more detail -- my biggest concern is average return / maximum drawdown for a reward / risk ratio.
ReplyDeleteSo for a year it would be:
SPY = 17.53% reward / risk (the risk is greater than the reward)
Sector = 128.41% reward / risk (the reward is greater than the risk).
Hi Tim,
ReplyDeleteAvg Return vs Max DD is indeed an industry standard metric. My rationale for Model vs XL* detailed analysis was to get an understanding of how well the model selects from the 9 possibilities at each decision point (I do acknowledge the timing of these decision points is co-dependent on the current selection so it might be that a somewhat different analysis is called for). Nevertheless, I feel that scrutinizing the decision making ability of the model is prudent. For example, if you find that the performance of the selected XL instrument is ranked below median (out of the 9 XL choices) in over 50% of the segments, thereby making the overall return outperformance highly concentrated (in time), it would be a reason for concern.
Wouldn't the metric in that case simply be Sector Model return vs Median return of all sector ETFs? That would be simple to do. But selecting the best every time wouldn't tell you if it was below median.
ReplyDeleteTo clarify, what I'm suggesting is to explore how consistent is the outperformance of the model by using a statistically-significant data set of XL segment choices. More important than comparing against the best or even median selections, would be IMHO to plot the histogram of model ranks (in each segment, relative to the rest of the sectors) as well as look at lengths of out-performance/under-performance periods. That analysis should demonstrate whether the overall outperformance can be reasonably trusted to endure (as opposed to be over-optimized/curve-fitted to a particular market environment).
ReplyDelete