First, auxiliary lines of the legend are a great help.
Vertical: in your exemplary case I wouldn't bother about vertical lines, since the step is pretty clear — it's 0.5, but if you look at the Dow Jones chart, you can see they could be really helpful for the user to follow the days.
With candlesticks chart vertical lines of legend can interfere with prediction sticks and complicate their readability, so they should be as subtle as possible (I typically go with a shade of grey, something between #f3f3f3 and #cecece).
Horizontal: are simply vital. Without them the viewers can only guess roughly about the value of a stick or a candle somewhere in the middle of the chart.
How to use and design the legend lines effectively: Tufte gave an exhaustive concept of the legend design, and you can google it by the "data-ink" keyword.
Second, two-colored candlesticks are more readable than the combination of colored and stroked ones.
Third, visual clutter.
- If you compare your chart to the other examples, you will see that
other charts have more air — more "white" space in them, so that data doesn't fill
all the coordinates. Allow some margins between the data and the
border, it will help decluttering.
Coloring sticks will also help to de-clutter.
There should be some space between the sticks, either vertically (like in Don Jones) or horizontally. Avoid making chart look like a bunch of a parallel black lines.
- Legend in your example has too high contrast and therefore is too aggressive. Again, data ink concept will help to work this out.
Fifth, the thickness here works as an aid (albeit non data-transmitting) to visual differentiation between the candle and the stick. Right now the width contrast on all four charts is pretty similar and perfect for recognition: stick is 1 px, candle is cca 20 px. If you play with the candle width, you will see that anything less than 20, like 10 or 5 px, is too close to stick, and anything more, like 30 or 50 px, is too much and thus "kills" the stick, by rendering it visually unimportant.