Statistical and Creative Analysis of Small Qualitative Dataset

The Central Bank (CB) meets bimonthly (every other month) to make decisions on whether to hike (increase), retain (hold) or reduce the monetary policy rate (MPR).

Ahead of their meeting, I try to gauge and ‘forecast’ what the likely outcome (increase/retain/reduce) will be (and by what extent).

After the meeting I make comparisons of what I ‘forecasted’ and what the actual outcome is.

I also look at what other analysts in the economy have predicted regarding the outcome of the meeting with respect to the Central Bank’s decision on MPR.

I have been 'predicting' the outcome of the Central Bank meeting for over two years now and I would like to ‘scientifically’ measure my ‘success rate’ taking into consideration:

1. How many times I got it ‘exactly’ right i.e say I predicted the CB will increase the MPR by 100bps and they increase it exactly by that amount.
2. How many times I was directionally correct in my forecast i.e say I predict the CB will increase MPR by 100bps but they actually increased it by 150bps
3. How my forecast stacks up to that of other analysts whose opinion I sample.
4. Finally, there is a slight twist which can be seen from the attached table. In some cases, my ‘prediction model’ actually ‘indicates’ that all things being equal, the CB will take a particular decision, but due to some ‘extraneous’ factors (such as comments made by CB governor) I know that only the ‘behaviour’ of one variable (among the lot of variables I consider) will really matter. So in such instances, I actually decide to go against what my model would necessarily have indicated.

I need someone with a good statistical/analytical background to come up with a way to calculate and come up with a score to measure how well overall I have performed taking into consideration all the aforementioned points.