6. Can GDP-weighted stock momentum predict bond ETF prices?

This post examines the model's performance on bond ETFs and on IWM- the small cap equity ETF.
​The bond ETFs are:
1) LQD, investment grade bonds based on S&P, Moody's and Fitch ratings.
2) HYG, high-yield low credit rated bonds.
3) TLT, 20-year treasuries
4) EMB,  international sovereign bonds.

SPY's dollar trading volume overwhelms all others; it's 4 times Amazon's and Apple's which have the highest dollar volume for stocks. This is why trades from me and you won't have any effect. Dollar volume of the bond ETFs is small relative to SPY but it's larger than many stocks. Also they're usual bid-ask spread of a penny gives them liquidity. The analysis will show highly-rated bonds (LQD) are an alternative to SPY if one wants lower risk - and lower reward.

Besides trading dollar volume the table shows SPY-bondETF correlation which explains the model's  ability to predict bond ETF prices.

ETF or Stock	Price	Avg Daily Volume	Avg $ traded/day		Daily Correlation ​with SPY	Model Day ​Afternoon Correlation ​ with SPY
SPY	250	65,000,000	$16.0 billion
Apple	160	23,000,000	$3.7 billion
Amazon	1000	4,000,000	$4.0 billion
IWM (small-equities)	140	22,000,000	$3.1 billion		.87	.86
TLT (20-year treasuries)	125	8,000,000	$1.0 billlion		-.35	.08
LQD (BBB or higher bonds)	120	6,000,000	$0.72 billion		-.07	.36
EMB (International bonds)	115	2,000,000	$0.23 billion		.37	.48
HYG (BB or lower bonds)	90	11,000,000	$1.0 billion		.67	.69

Let’s compare the model performance of these ETFs to SPY’s performance using Monte Carlo simulation which offers an intuitive analysis of statistical significance. GDP-momentum and other factors in my model selected 131 out of 1200 days for a predicted SPY rise. Monte Carlo randomly selects 131 out of 1200 days then compares SPY’s performance on these days with the model days performance - and repeats,..., repeats this process. With 1,000,000 trials, statistical significance at the 1% level requires that less than 10,000 of the random selections perform better than the model’s selection.

Results for SPY are astounding. None of the one million random selections beat the model.

​The Monte Carlo column shows that model performance of equity small caps (IWM), of high-yield bonds and of international bonds are almost as powerful as SPY. Small caps, and high-yield bonds, are 87% and 68% correlated with the SPY so since the model predicts SPY, it's not surprising that  it powerfully predicts correlated ETFs. Foreign bonds, EMB, are 37% correlated with SPY so its return is a bit lower but  still powerful. 

(Now one may challenge HYG’s performance with a Win Ratio of 65.4% being equivalent to SPY whose win ratio is 74.0%. With both not even one in a million Monte Carlo trials could beat the model. One explanation is SPY's starting advantage. On all days its WR was 56.1% while HYG’s was 51.2%. The third last column showing average return divided by standard deviation, a risk measure, similar to the Sharp Ratio, provides another explanation . Here HYG's value (.48) equals SPY's paralleling the Monte Carlo results.)

The giant surprise was LQD's impressive performance even though it's negatively correlated with SPY  at minus 7%. On model-selected days, however, the afternoon correlation was positive 36% ! 

TLT is a mild surprise. Although it's the only ETF whose afternoon return isn't statistically significant its WR did increase from 50.7% to 55.7%. It's afternoon correlation with SPY on model days changed similarly to LQD. Their daily correlation is minus 35% yet on model days the afternoon correlation was positive 8% !

How to explain LQD and TLT? One speculation (and I mean speculation) for LQD rising in the afternoon on model days is that GDP-momentum causes interest rates to remain steady as shown by TLT's average increase of .031 (not drop as negative correlation of SPY-TLT would imply)  and credit risk to lessen. I'd appreciate options of economists and other readers of these notes.

ETF	All 1200 Days ​Min	All Days Daily ​Avg	All Days Std Dev	All Sharp Ratio: Avg/ StdD	All Days ​Win Ratio ​(WR)	​	Model ​131 ​ days ​Min	Model days ​daily Avg	Model days ​Std Dev	Model Days Sharp Ratio ​Avg/ StdD	WR	# beating model in 1 million Monte Carlo trials
SPY	-3.64	.016	.46	.03	56.1		-0.82	.206	.43	.48	74.0	0
IWM	-2.52	.026	.58	.04	54.4		-1.22	.220	.60	.37	68.0	67
HYG	-1.59	.012	.22	.05	51.2		-.82	.125	.26	.48	65.4	0
EMB	-1.49	-.004	.20	< 0	51.2		-.53	.083	.28	.29	60.9	3
LQD	-1.24	.002	.18	.01	51.4		-.46	.067	.24	.28	63.8	55
TLT	-1.82	-.005	.39	< 0	50.7		-1.02	.031	.41	.07	55.7	157,700