Dividend Yield focused strategies have become increasingly popular among investors seeking to enhance the cashflow and income generated by their portfolio. Much of the interest in this strategy is driven by a lack of sufficient investment opportunities in bond markets. Simply put, bond yields are far too low to warrant meaningful attention from income driven investors. 

Long/Short Portfolio
Typically, a portfolio with such large dividend yield comes with substantial risk. In order to mitigate this risk we'll use both long and short positions. This allows us to hedge out a significant portfion of this risk. 

We utilize the following assumptions for constructing our portfolio:

• Leverage: No more than 5:1 leverage will be allowed. Note that this level of leverage can only be achieved with a Portfolio Margin account. If you don't have a PM account, see our list of Portfolio Margin brokers to find a suitable broker.
  
• Min/Max Individual Stock Weighting: no more than 3.5% of the portfolio can be invested in any single stock (anything greater would result in far too much asset-specific risk).
• Portfolio Construction Method: Mean variance optimization is used to ensure the portfolio minimizes risk while maximizing dividend yield. The risk model used in this optimization attributes risk to style and sector factors. This is similar to models available from commercial vendors such as Barra, Northfield and Axioma.

Below you will find an analysis of the long/short portfolio, as well as the composition of the portfolio itself. Figure 1 shows that the portfolio has roughly the same risk as the S&P 500. You can also see how the portfolio compares to other portfolios of interest.

Figure 2 shows that there are 168 positions in the portfolio: 92 long and 76 short. The total risk is shown and is also decomposed across the long and short sides. The risk displayed here is quoted as the annualized standard deviation of returns. Note that this risk in the "Longs" and "Shorts" columns is additive. That is, the risk on the long side plus the risk on the short side equals the Risk in the "Total" column. If you look at the "Shorts" column, you will notice that many of the numbers are negative. This tells out the short positions are acting as a hedge to the long positions and therefore reduce risk.   

Risk is also attributed to "Common Factor" risk and "Asset Specific" risk. Common Factors drive returns across multiple assets and thus are the source of correlation between securities (think sector/industry risk and the so called "style" factor risk such as value/growth and size). In contrast, asset specific risk emanates from a the operations of an individual company and is diversifiable within limits (i.e. asset specific risk decreases as we add positions to the portfolio). As we can see, Common Factor risk dominates the portfolio at 14.63%, while Asset Specific Risk is 7.74%. To get to the Portfolio Total Risk, we add the square of the Common Factor Risk to the square of the Asset Specific Risk and then take the square root of this number.

Figure 3 shows the exposure and risk attributable to the common factors. Focus your attention on the last column which represents the amount of risk contributed by exposure to the given factor. As we can see, most of the risk is driven by exposure to three of the common factors: Utilities Sector factor, Services Sector factor, and the Growth/Value factor. The table also shows that some of the factors act as a hedge and thus end up reducing risk. For example, exposure to the Health Care factor reduces risk by 1.442%

Below is a table containing the portfolio's positions assuming a portfolio net equity value of $1,000,000. The share quantity can be scaled up or down to accommodate the actual amount available to be invested. For example, if you only have $100,000 to invest then scale down the positions by a factor of 10. 

Most of the columns in the table are self explanatory, however, the "Component Total Risk" deserves further commentary. The Component Total Risk represents the risk emanating from that specific position considering the portfolio context. A positive value means the position adds to risk. A negative value means the position decreases (i.e. hedges out) risk. The sum of these values is equal to the "Portfolio Total Risk" show in Figure 2.

The last two blog posts focused on quantifying news stories to shed light on Earnings news and M&A news. Today, we'll carry on with this theme to look at "Recession" news. Again, the idea here is to collect news articles on S&P 500 stocks and then compute the percentage of that overall news flow that has to do with "recessions".  

The charts below plot recession news (see top chart) and the SPY ETF (i.e. S&P 500 ETF) on the lower chart. 

The most salient point we can deduce from these graphs is that recession news spiked in Aug 2011 to around 0.35% of the overall news flow. This massive spike corresponded to the sharp market sell-off. Then, as recession fears mitigated, recession news flow dissipated, and the market began it's 2012 rally. 

As investors, we have some obvious questions about this metric: 

• Does it have predictive value?
  
• Does the market follow the news or does the news follow the market? 
  

I don't have a long enough data history to render a definitive conclusion. What's clear however is that news chatter around "recession" certainly doesn't help the market. 

Looking at the more recent data, it appears that recession news is nearing a post spike low. This seemingly should bode well for the market going forward. 

Apropos of my previous blog post discussing quantifying news stories, I present an additional news metric that provides food for thought.

The graph below represents M&A news chatter activity through time, with the Y-axis representing M&A news as a percentage of the overall news flow. The graph shows that M&A news peaked in Q1/Q2 2011 (at around 1.3%) and has been declining ever since.

Like many traders I watch CNBC and Bloomberg TV religiously. Keeping abreast of breaking news, market happenings and listening to the talking heads espouse and debate their opinions - what could be better, right? While you may not always agree with the viewpoint being promulgated, understanding the merits of the argument has value in and of itself. 

Fast forward a few hours later and the market had taken a sharp turn to the negative. Once again, I tuned to CNBC to see what had changed to warrant this new market direction. Lo and behold, the only thing that had changed was their conclusion! The same narrative, the same compelling arguments, the same supporting data, but this time used to sustain the exact opposite hypothesis - that the market was surely destined to be headed into negative territory. 

Now, I'm not a prosecutor and the financial markets are certainly no court room. Market prognosticators and commentators have the right to change their opinion should they find themselves in err. What I find so irksome however is the depth of their conviction. The situation certainly called for an iota of humility given how incorrect the previous assessment was.

What's the lesson to be learned here? Is it reasonable to castigate the story tellers when - after all - it's their job to tell us a story. No, what I lacked were the tools to independently determine the value of news information. I needed to attain a clearer, more unfiltered view of what the news (including opinions and editorials) was saying. So, I did what seemed logical to a programmer like myself, I decided to build something. The next few weeks were spent coding software that would absorb news information from across the web and perform semantic analysis to garner meaning from the jumble of words. 

Was there a larger story to the news flow then simply what could be determined by reading a handful of articles? Could my software divine the meaning of this news - if not at the individual story level then at least at the aggregate level? If so, would this information be of value?

Several months after I finished the project the financial crisis settled down and I all but forgot about my news analysis program. Luckily, the code has been running continuously on one of my servers, dutifully pulling news data from across the web and crunching numbers. I recently checked in with the program to see what it had to say.  

One of the metrics tracked by the program is news related to earnings. Figure 1 show the ratio of "good earnings news" to "bad earnings news", computed by classifying earnings news as either "good" or "bad" and then computing the ratio of the two. A simple example of this classification method is as follows: if the news story was about a company beating its earnings guidance then the article would be classified as "good earnings". In contrast, an earnings miss would result in a "bad earnings" classification. 

This "Good vs. Bad" earnings metric can be viewed as a guide to what type of earnings information dominates the news flow. The day-to-day metric shown in Figure 1 is quite volatile and so the moving averages show in Figure 2 provide for easier interpretation.

What's the key piece of information being communicated here? 
Well, the chart tells us that positive earnings news peaked in Q3/Q4 2010 and has been declining ever since. In fact, we're close to a three year low in S&P 500 earnings as far as the news chatter is concerned.

The earnings metric is but one of many metrics currently being tracked by my news program. While the value of this information is still to be determined, it certainly does communicate novel information and sheds a quantitative light on the qualitative world of news. While I haven't given up my CNBC just yet, I'm now able to check a commentator's opinion of the news against the quantitative reality.