michalkokta

Pound the Knees, Steven

The Toronto Blue Jays have recently traded for left-handed pitcher Steven Matz. Matz’s 2020 was a year to forget – join the club Steven – but FanGraphs projects him to slide into the bottom of the Jays starting rotation, and pitch about 132 innings this year. Let’s take a look at who Matz is as a pitcher, and why a change in fastball location is something the Jays coaching staff might consider.

Matz pitched only about 30 innings last year, so in the interest of sample size, I will also be using statistics from 2019 and 2018. Here is what Matz’s last three seasons looked like:

There is a lot of moving parts in there, but I’d like to direct your attention to the fastball ratings. Specifically, how Matz’s fastball, in the past three seasons at least, had an above average velocity, and a below average raw spin rate. That in and off itself doesn’t make a pitch “good” or “bad,” but might hint at its effectiveness in different parts of the zone.

Matz stopped using his four-seamer in 2018, and now only throws one type of fastball – a sinker. The sinker is his bread and butter pitch; he has thrown it over 50% of the time in each of the past three seasons. It sits in the 93-95mph range, with a spin rate that has been increasing, but was still under 2,300 rpm in 2020. And there just might be a way for Matz to get more out of it.

I’m not sure how familiar you are with Bauer units. Yes, that Bauer, the reigning Cy Young award winner. The formula for Bauer units is really simple; it’s just the spin rate of a pitch in rpm, divided by its velocity. Why would we want to do that? It allows us to compare and classify pitches as high spin, or low spin, while normalizing for velocity. Fastballs thrown at a higher velocity tend to spin at a higher rate, and Bauer units allow us to remove velocity from the equation when comparing how “spinny” a particular pitch is.

Why is that important? According to the linked Driveline article, the average MLB fastball is around 24 Bauer units. Fastballs with higher Bauer units tend to be more effective up in the zone; these are the fastballs that seem as if they are “rising” to the hitter, and hitters struggle to square them up. The poster boy for a higher Bauer unit fastball is – ironically, if you know their UCLA and even pro history – Gerrit Cole. In 2020, Cole averaged 2,505 rpm on his four-seam fastball at 96.7 mph. That’s 2,505/96.7 = 25.90 Bauer units. And here is Cole’s four- seam fastball heat map from 2020, with a clear preference for locating his four-seamer up in the zone.

This brings us back to Steven Matz. Here is a table of his sinker Bauer units from the past three years.

Year	Sinker RPM	Sinker Velocity	Bauer Units
2018	2,069	93.4 mph	22.15
2019	2,102	93.3 mph	22.53
2020	2,230	94.5 mph	23.60

Courtesy MLB Statcast

The sample size for 2020 is not great, but regardless, Matz’s sinker is a low spin offering and should therefore be more effective low in the zone. Let’s see what the batted ball data says.

The above is only for the sinkers in the three seasons 2018 – 2020. When Matz throws the sinker in the bottom third of the zone, and contact is made, you can hear it. We have exit velocities of 93, 95, and 97mph. However, after you hear the loud contact – with the advances in fielder positioning, shifts etc. – the next thing you hear is probably the smack of the ball landing in an infielder’s glove. That’s because the launch angles in the bottom of the zone are all below 10 degrees, with the lone exception being when Matz misses down to his glove side. Do the outcomes line up with the underlying metrics?

If Matz stays away from the middle of the plate, he has more success down in the zone. The wOBAs are decreasing as we travel down in the zone to Matz’s glove side: .342, .302, .175 from top to bottom – as well as to his arm side: .349, .250, .151. Down to his arm side seems to be his most effective location: wOBAs of .151 in the zone, and .145 on the chase, driven by launch angles of 3 and -6 respectively.

With all that out of the way, let’s take a look where Matz has actually been throwing his sinker in the past three years.

Well, what do you know – Matz has mostly been working his sinker up in the zone. Remember that location with low wOBAs and launch angles from the pictures above? Matz has gone that way hardly at all. The reason, of course, might be that if Matz goes down and to his armside, he catches the batter off-guard and that’s why he’s been successful when he has gone down there when he has. Yet I wonder if this is something the Jays might have picked up on? We’ve all read the stories of Astros targeting high spin guys in trades in their World Series years, and getting them to work up in the zone more. Gerrit Cole is a prime example of that. Could the Jays have seen a potential adjustment to be made and bought low on a pitcher coming off a disastrous 2020?

There is obviously more to a pitcher’s plan of attack than just location of one pitch; how a pitcher’s full arsenal of pitches compliment each other, pitch tunneling, all of those are important. I could be way off base, but now I’m curious whether Matz’s sinker will stay more or less the same in 2021, or if we’ll see him try to keep it lower in the zone a little bit more.

Teoscar Hernandez: Is It Real?

As a baseball fan, who recently relocated to Toronto, I feel like I should have at least a passable knowledge of what’s going on with the Blue Jays. My interest in baseball developed while living in Texas, and I have been a supporter of the Texas Rangers ever since – you might know that the Rangers and Blue Jays have a bit of a history. As I was browsing the various Statcast leaderboards from last year, I came across Teoscar Hernandez and his progress in 2020.

Statcast formulates the xwOBA metric based mostly on launch angles and exit velocities of a particular hitter’s batted ball events, and uses that data – along with walks and strikeouts – to predict what the hitter’s wOBA “should” be. And based on xwOBA, Hernandez was the third most improved hitter in baseball in 2020. The coach in me wanted to look under the hood and see, first, what drove Hernandez’s progress, and second, how sustainable it was and whether there were any red flags or possibilities for regression.

Starting with the various projections for 2021, Steamer and ZiPS are split regarding Hernandez’s offense:

In 2019 Hernandez’s offensive output was worth 1.8 runs above average, and that number spiked to 12 runs in 2020. While both Steamer and ZiPS see Hernandez as a well below average fielder, ZiPS seems to mostly buy Hernandez’s improvements, projecting him to be worth 9.9 runs above average in 2021. Steamer, on the other hand, is skeptical, and expects Hernandez’s offense to regress to 2019 levels. So how did Hernandez go about increasing his offensive metrics in 2020?

Looking at the table above shows that the driver behind Hernandez’s xwOBA increase was the fact that he simply hit the ball harder: his average launch angle stayed the same at 15.3 degrees, while his exit velocity metrics increased. He squared the ball up more often, going from 11.7% to 18.0% barrels; as a result, his hard hit percentage (the percentage of balls that left his bat at 95mph+) jumped to 53.1%, and his average exit velocity increased by about 2mph.

Breaking down Hernandez’s hitting statistics by pitch type reveals that his increase in both xwOBA and exit velocities was achieved almost exclusively by punishing fastballs.

Hernandez’s xwOBA on fastballs was an astronomical .486, driven by a 20 degree launch angle and an average exit velocity of nearly 97mph. However, there is a problem with putting all your eggs into the fastball basket.

While being an elite fastball hitter, Hernandez has struggled with hitting breaking and offspeed pitches. In the past three years, his xwOBA on breaking balls is yet to surpass .300, and he has been swinging and missing at them at a rate of about 40% or worse. He has been able to improve his whiff percentage on the offspeed stuff, but has been pounding those pitches into the ground with an average launch angle of 7 degrees. The worrying part, for Hernandez anyway, is that pitchers have noticed, and he has been seeing fewer fastballs: from about 57-58% in 2018 and 2019, to about 53% in 2020. The decrease in fastballs has been offset by an increase in the number of offspeed pitches thrown to Hernandez, but given the data above, I would not be surprised if he starts seeing a larger share of breaking balls in 2021.

Moreover, Hernandez so far hasn’t show the ability to get on base consistently in a different way than pounding the fastball.

His 2020 strikeout rate of 30.4% was in the 12th percentile for all batters, while his walk rate of 6.8% was in the 24th percentile. Walks and strikeouts are certainly not the end-all-be-all, but Hernandez is well “above” the MLB average in terms of his whiff percentage – 36.0% for his career as opposed to 24.5% MLB average, caused in large part by his below average zone contact percentage: 71.5% for his career as opposed to 82.8% MLB average.

Baseball is a game of adjustments, and it will be interesting to follow Hernandez’s 2021 season. I would not be surprised if Hernandez saw less than 50% fastballs going forward, and it will be up to him to show the pitchers that he can handle the breaking ball in the zone. Given the league-wide trend of overall decreasing fastball usage – and Hernandez’s limited contributions on defense – learning how to do damage on something other than fastballs is a critical skill for Hernandez’s future as a major league contributor.

Second Serve +1 = 3

Knowing that even following a second serve, the majority of points in men’s professional tennis are contested in the 0-4 shot range, what are some of the ways in which players could go about increasing their effectiveness in these short rallies? As a server, you want to maximize the number of “1” and “3” shot rallies in that 0-4 range, i.e. the rally lengths that lead to you winning the point (remember that rallies of an odd length mean a point for the server, while rallies of an even length lead to a point won by the returner). Are there any particular areas of the service box you should target more often with your serve? Furthermore, let’s say you’d prefer your first stroke after the return to be a forehand. How does that change the equation?

Once again, I decided to use the dataset of all individual points played in the 2020 US Open men’s singles matches, compiled by Jeff Sackmann. Specifically, I wanted to focus on points starting with the second serve, and a rally length of 3. (The few lines of code needed for the article can be found here.)

Second Serve Rally Lengths, 2020 US Open Men’s Singles

There are two reasons why I think this particular scenario is important. The first is simply that it was the most frequently occurring rally length in dataset (see above). The second reason is slightly more tactical. If I wanted to maximize the rally length of 1 after a second serve, that means I don’t want the serve to come back. I need to either hit it harder, with more spin, risk more, or be more unpredictable with its location. On the other hand – maximizing the rally length of 3 – I want the serve to come back. I just want it to come back to where I want it, so I can do damage with my “+1.” I don’t need for the serve to be the dagger and take a lot of risk with it; I only need it to set up the rest of my game.

Let’s start with the 30,000ft view. Overall, there were 6,541 second serve points in the dataset, for which the serve direction was tagged. The direction was assigned one of five values: Wide, Body-Wide, Body, Body-Center, and Center. It’s important to mention that the 6,541 points are both deuce and ad points combined, as well as righty and lefty servers combined. Here are the direction percentages for all second serve points:

Second Serve Directions: Overall

Direction	W	BW	B	BC	C	Sum
Serves Hit	688	1,700	1,214	1,824	1,115	6,541
% of Serves Hit	10.52%	26.00%	18.56%	27.89%	17.05%	100%

We can learn two things from the table: one, the body serve reigns supreme. When we combine all three of the body serve categories, we see that more than 7 out of every 10 second serves were aimed at the body of the opponent. The body serve is especially effective when the returner wants to take the return early, or on the rise – right around the baseline or inside it. It will lose some of its bite when the returner stands 2+ meters behind the baseline and has time to move and make space for the return. Second, the players are apprehensive about opening up the court for the opponent, going wide with their second serve only about 10% of the time.

With that general overview of the second serve direction patterns, let’s zoom in on the second serve patterns that yielded the rally length of 3 most often. There were 1,577 second serve rallies of length 3 in the datatset, tagged with the corresponding serve direction. The breakdown is as follows:

Second Serve Directions: Rally Length = 3

Direction	W	BW	B	BC	C	Sum
Serves Hit	173	408	315	445	236	1,577
% of Serves Hit	10.97%	25.87%	19.97%	28.22%	14.97%	100%

Looking at both of the above tables combined, it doesn’t seem that serving into one particular area of the box resulted in more 3 stroke rallies than expected; the percentages are all basically the same. The biggest difference was in the “Center” location, where players directed 17.05% of their second serves, but that location yielded about 15% of three stroke rallies. That 2% difference is probably due to the fact that after a second serve down the center, both the server and returner are positioned near the middle of the court, not dissimilar to what you would see in the warmup. It is harder to generate any attacking angles from that part of the court, and we’re likely to see a longer baseline rally as a result.

If there aren’t any obvious areas of the box to serve into to increase the likelihood of the 3 stroke rally by themselves, are there any areas that increase my chances, as a server, that my “+1” will be a particular stroke? I would argue that the majority of players would prefer that first stroke after the serve to be a forehand.

Short of watching every point and tagging the first stroke hit by the server following the return, we’ll need to use a proxy. If a specific rally in the dataset ended with a winner, the point is tagged with either a ‘F’ or a ‘B’ – forehand or backhand – based on the type of shot that resulted in the winner. What we have below are second serve points, of rally length 3, that ended with a forehand winner. Overall, there were 373 of those points in the dataset, with the distribution as follows:

Second Serve Directions: Rally Length = 3, Winner = ‘F’

Direction	W	BW	B	BC	C	Sum
Serves Hit	65	100	61	94	53	373
% of Serves Hit	17.43%	26.81%	16.35%	25.20%	14.21%	100%

The thing that stands out in this table is the column with the “Wide” serve direction. In the tables above, we saw that about 10.5% of all second serves were hit into the wide section of the box, and 11% of all second serve 3 stroke rallies started with a wide second serve. But when looking at second serve 3 stroke rallies finishing with a forehand winner, that number jumps to over 17%.

Obviously looking at a small subset of points – the ones ending with a winner – has its limitations. For example, we are not considering rallies that ended with a forced error by the returner on stroke number 4. However, I think that there is something to pay attention to even with the relatively small sample size.

We saw from the data how infrequent second serves out wide are – only about 11% of second serves were hit out wide in the dataset. I would guess the percentage would be even smaller than that in just the deuce side of the court. In the ad side, the kick out wide tends to be the easiest topspin serve for a right handed server to hit, and the wide “slider” in the ad is the bread and butter of lefties. So if a player decides to hit a second serve wide in the deuce, he often catches the opponent by surprise, resulting in a weaker return. And on that weaker return, the server has time to move around his backhand if he needs to, and get on offense with his forehand. Moreover – just by the geometry of the court – it is harder to direct a return from the wide section of the deuce side into the ad side of the server: the net is higher, and the court is shorter that way. As a result, the return will probably come back through the middle, or cross court – allowing the server to “cheat” that way in preparation for his +1.

If you are a player yourself, or maybe coaching a player, and you would like to increase the chances of crushing a forehand following your second serve, don’t forget about the wide target. Especially in the deuce side.

Aaron Nola Will Make You Question Yourself

In one of the later chapters of The MVP Machine, the authors describe a working relationship between a professional baseball player (an unnamed position player) and a writer at an “analytically inclined” baseball website. The player felt that his club’s advance scouting data wasn’t granular enough, and asked the writer to supplement the information with more detail. The writer summarized that the player was basically looking at three things: “Am I squaring up the ball? Am I swinging and missing? Am I swinging at strikes?”

That last question got me thinking. As a pitcher, it is rarely a bad idea to have batters look at called strikes, and swing at balls. Which pitchers, in 2020, were particularly effective at doing just that? To make that determination, I looked at Statcast data for all pitchers, who threw at least 60 innings in 2020. Specifically, I looked at their outside-zone swing rate, their zone take rate – calculated as just (1 – zone swing rate) – and took the average of the two. Note that this analysis completely omits of what happens if contact is made with the ball. We’re merely interested in taking strikes, and swinging at balls. (If you’re interested in the Statcast query and the few lines of code for this, click here.) The top ten was as follows:

Name	Oz_swing_%	Z_take_%	Avg
Aaron Nola	0.366	0.397	0.3815
Zac Gallen	0.294	0.405	0.3495
Kenta Maeda	0.393	0.306	0.3495
Brady Singer	0.271	0.427	0.3490
Shane Bieber	0.344	0.353	0.3485
Jose Berrios	0.312	0.384	0.3480
Alec Mills	0.299	0.393	0.3460
Dylan Bundy	0.291	0.396	0.3435
Marco Gonzales	0.303	0.377	0.3400
Adam Wainwright	0.301	0.376	0.3385

Courtesy of baseballsavant.mlb.com

Aaron Nola was in a league of his own in this made-up stat in 2020. The difference between him and Kenta Maeda with Zac Gallen is as big as the difference between Maeda/Gallen and the 22nd player on the list. Let’s take a look at how Nola goes about creating uncertainty in the batters’ minds. We’ll look at what happens before he even throws a pitch, and once a pitch leaves his hand.

Before

Imagine you’re a major league baseball player, and you are about to face Aaron Nola. Starting with the 30,000ft view, which pitches does he throw, and how often? In 2020, the breakdown looked like this:

Pitch Type	#	#RHB	#LHB	%
Changeup	314	164	150	27.4
Curveball	306	156	150	26.7
4-Seam FB	290	115	175	25.3
Sinker	238	148	90	20.7

Courtesy of baseballsavant.mlb.com

Nola throws four pitches, and relies on three of them about equally. He used the sinker the least, but still threw it about a fifth of the time. Regardless if you’re batting righty or lefty, you have to respect the curveball, and the changeup. If you’re batting righty, you might get the sinker more often, and if you’re a lefty, maybe the four seamer. But we haven’t learned much.

What about by a particular count?

Except for the 3-0 fastball, maybe look for the 2-1 changeup? But other than the 3-0 count, we see every pitch being thrown in every count. And there really aren’t any big circles in the picture – no particular pitch dominates a particular count, with the exception of 3-0.

OK, sounds good coach. We know we will get one of four pitches, but very little information on which one might be coming when. Maybe we’ll learn something in the 0.4 seconds we have between a pitch leaves Nola’s hand and we have to decide whether to swing or not?

During

Now you’re standing in the batter’s box. Nola goes into his motion, and delivers the pitch.

According to Baseball Savant, Nola has a “very consistent release point.” The changeup comes in from about 4’8″, the fastballs from 4’9″, and the curveball from about 5’1.” What happens during the release is even more fun.

Let’s compare the two fastballs first:

Fastball	MPH	Active Spin	Spin Axis	Horizontal Break	Vertical Break
4-Seam	92.8	98%	1:45	13.9in	18.1in
Sinker	91.7	96%	1:45	17.4in	24.4in

Courtesy of baseballsavant.mlb.com

That’s two different types of a fastball, coming in at about the same velocity, thrown from the same release point, and spinning around the same axis at release. But one of them has about 3.5 inches more arm side run, and more than 6 inches of additional vertical drop. I would imagine it is nearly impossible to tell them apart before one needs to make a swing decision.

If it’s hard to tell the two fastballs apart, what about the curveball?

Pitch	MPH	Active Spin	Spin Axis	Horizontal Break	Vertical Break
Curveball	78.6	86%	7:45	15.1in	54.8in

Courtesy of baseballsavant.mlb.com

The curveball in and off itself averages 4.1 inches more vertical drop than other curveballs thrown at comparable speeds from comparable release points. What amplifies its effectiveness is how it compliments the fastball. Or in Nola’s case, two fastballs.

I will not attempt to explain spin mirroring on here, Michael Augustine does a much better job of it than I ever will. But notice that the spin axes for the fastballs (1:45) and the curveball (7:45) are exactly 180 degrees apart. In essence, the fastballs and the curveball are spinning along the same axis; the difference being that the fastballs are rotating “backwards” through the air because of the backspin on the ball, while the curveball is rotating “forwards” because of the topspin on the pitch. And for a batter to differentiate between topspin and backspin on a pitch in a split second is extremely difficult. (Seriously, go read the article. Well worth your time. There is a great animation demonstrating this point in there too.)

Alright, so what do we have so far? Three pitches, extremely difficult to tell apart when they leave Nola’s hand, and can be thrown in any count. Great. Let’s play a little game.

We have two fastballs, coming in at 92.3mph on average, and a curveball coming in at 78.6mph. If we were to add another pitch, to upset a hitter’s timing further, how about we have it average about (92.3 + 78.6)/2 = 85.45mph. Also, let’s have the vertical break be halfway between the vertical break of the curve and the vertical break of the sinker? Around 40 inches?

Pitch	MPH	Active Spin	Spin Axis	Horizontal Break	Vertical Break
Changeup	84.9	99%	2:30	14.5in	35.5in

Courtesy of baseballsavant.mlb.com

40 inches of vertical drop on the changeup would be Logan Webb/Devin Williams territory. Nevertheless, the changeup fits perfectly with the fastballs and the breaking ball in terms of speed. It is released from a slightly lower spot than the fastballs, but has about 10 more inches of vertical drop compared to the sinker. In 2020, opponents were just pounding it into the ground, slugging .296 on the pitch with a -6 launch angle.

To summarize, Nola has 4 pitches that complement one another extremely well, and the ability and willingness to throw any one of them in any count. His unpredictability resulted in uncomfortable at bats for the position players in 2020, who were more prone to look at strikes and swing at balls against Nola than against any other starting pitcher.

2020 US Open Rally Lengths

It has been fairly well established that modern tennis is a game of short rallies. Rallies shorter than 4 shots – i.e. serve, return, one additional shot in the court by the server, and one additional shot in the court by the returner – account for about 70% of all rallies in a match. About 20% of the points then tend to fall into the 5-8 range, and only about 10% of the points will be the extended rallies we see in the highlight reels.

With that being said, I wanted to tackle the following two questions:

Are the rally length distributions similar regardless of whether the point started with a first serve or a second serve? In other words, are points that start with a second serve longer? And if so, longer by how many strokes?
Knowing that the first serve winning percentages can climb upwards of 80%, while the second serve winning percentages hover around 50%, how exactly do returners make up that 30%? Specifically, do the returners tend to win their points off of second serve returns in short rallies, quickly following the return? Or are they winning extended baseline battles?

To answer those questions, I looked at the 2020 US Open men’s singles point-by-point statistics, collected by Jeff Sackmann. The dataset contains 27,395 individual points tagged with features such as rally length, serve direction, or distance covered by both players, and can be found here. (If you want to see the code for the analysis below, click here).

Below are the rally lengths broken up by whether the point began with a first or a second serve. Note that I’m ignoring rallies of length 0 – double faults.

Rally Length Following a First Serve (2020 US Open Men’s Singles)

Rally Length Following a Second Serve (2020 US Open Men’s Singles)

Three things immediately stand out:

Both graphs are significantly right-skewed, confirming the prevalence of the 0-4 rally length on both first and second serve points. Overall, there just weren’t that many rallies longer than 5 strokes, regardless of what serve was put in play.
However, looking within the 0-4 range itself, there is a difference. In the first serve graph, we see two spikes at rally lengths 1 and 3; intuitively that makes sense – both 1 and 3 stroke rallies are points for the server. In particular, a rally length of 1 is either and ace or an unreturned serve, while rally length of 3 is a “Serve + 1” pattern by the server, where the “+1” results in either a winner, or an error by the returner.
Looking at the 0-4 range in the second serve graph, we see a smaller contribution of the 1 stroke rally – i.e. second serves get returned more often than first serves – but the key here is the spike in the rally of length 2. A rally of length 2 is a point for the returner – serve in play, followed by either a return winner, or an error by the server on shot number 3. Also, we have further confirmation of the importance of the “Serve +1” patterns for the server, with the rally of length 3 being the most common following a second serve in play.

Here is a table outlining some of the same data:

Serve in Play	Rally Length (Mode)	Rally Length 25th Percentile	Rally Length Median	Rally Length 75th Percentile	0-4 Rallies	5-8 Rallies	9+ Rallies
1st	1	1.0	2.0	3.0	88.25%	7.83%	3.92%
2nd	3	2.0	3.0	4.0	81.66%	11.08%	7.26%

Were second serve rallies longer at the 2020 US Open? Yes, but with a caveat. The mode, 25th, 50th, and 75th percentile values were all higher for the second serve points. We also see a bigger proportion of the rallies going past 5 strokes on the second serve. With that being said, over 80% of second serve points were still contested in that 0-4 range. Interestingly enough, the decrease of about 6.5% in the 0-4 point length between the first and second serve points is about evenly distributed among the 5-8 and 9+ rally lengths; both increased in the neighborhood of about 3.2%.

Now that we know the answer to the question of “are second serve rallies longer” is “yes, but..,” let’s take a look at where points won on the return come from. To do that, we need to look at “even” rally lengths: remember that rallies of length 2, 4, 6 etc. all indicate a point won by the returner. Comparing the frequencies with which rallies of a given length occur on first or second serves will give us insight as to where the returners do most of their damage.

Serve / Rally Length	2	4	6	8
1st	16.67%	6.86%	1.84%	0.97%
2nd	26.68%	9.62%	2.91%	1.69%
Difference	+10.01%	+2.76%	+1.07%	+0.72%

Looking at the table above, it is clear that the rally length of 2 – i.e. serve in play, followed by a return winner, or an error by the server on stroke 3 – was the biggest difference maker for the returners. About 26.7% of all points that started with a second serve were of length 2, as opposed to only 16.7% of points that started with a first serve. This difference of 10% drops off dramatically once the rally gets to 4 strokes and longer, meaning the effect of the serve and return diminishes pretty quickly after the third and fourth stroke of the rally.

I always wondered whether the players with high second serve return win percentages were good returners, or simply good baseliners, who put the return in play and grind out long rallies. This data suggests the former: good second serve return win percentage will probably be driven by strong performance in the 2 and 4 length rallies, highlighting the effect of the return itself.

A simple takeaway from this exercise is as follows: as you sit down to watch the Australian Open in the next few weeks, you have two options. If you want to see tennis for what it is and watch the whole matches, it will be a lot of serving and returning and not much else most of the time. If, on the other hand, you are a rally aficionado, I would just wait for the highlights.

Diego Schwartzman’s Serve Adjustment

One of the reasons I love tennis is because of players like Diego Schwartzman. The outliers. The ones, who are “not supposed to be there.” As of January 4th, 2021, Schwartzman is ranked #9 in the world, despite:

Being listed at 5’7,” while all the other players in the top 10 are 6’1″ or taller

Not having a high profile as a junior – his career high ITF World Junior Ranking was #217 – while most of the other players in the top 10 were ranked #100 or better

Holding serve less than 80% of the time, while the other players in the top 10 are comfortably over that mark

It’s this last bullet point that will have to change, if Schwartzman is to keep ascending up the ATP rankings: he will have to develop his serve into more of a weapon. He already excels on the return: according to the ATP Stats Leaderboards, he was the 2nd best returner on the tour in the year 2020. Yet when looking at the serve leaderboards, he ranked 48th. With that being said, let’s take a deeper dive into some of Schwartzman’s serve metrics to see where the improvements could come from.

First, this is how Schwartzman’s serve statistics compare to the rest of the top 10. The rankings are as of January 4th, 2021, and all statistics are for the year 2020.

ATP Rank	Name	First Serve %	First Serve Win %	Second Serve Win %	% of Service Games Held
1	N. Djokovic	64.4%	75.2%	54.1%	86.3%
2	R. Nadal	64.4%	73.6%	58.6%	87.1%
3	D. Thiem	63.0%	73.8%	51.7%	83.7%
4	D. Medvedev	59.1%	76.8%	53.6%	83.4%
5	R. Federer	66.0%	74.2%	53.0%	86.1%
6	S. Tsitsipas	64.5%	76.1%	57.8%	88.4%
7	A. Zverev	68.0%	76.7%	45.9%	82.5%
8	A. Rublev	59.0%	78.6%	53.4%	85.2%
10	M. Berrettini	62.1%	80.2%	58.0%	91.0%
Mean (no Schwartzman)		63.4%	76.1%	54.0%	86.0%

9	D. Schwartzman	62.7%	64.7%	50.9%	72.9%

Data Courtesy of ATPTour.com

Looking at the columns individually, the first serve percentage is right in line with the rest of the top 10 – there is no need for Schwartzman to get more first serves in play. Similarly, the second serve win percentage is not an outlier either. Schwartzman is about 3% below the rest of the top 10, but his second serve win percentage is better than Zverev’s, and close to Thiem’s, for example.

The reason why Schwartzman holds serve only about 73% of the time, while the rest of the top 10 is at 86%, is that Schwartzman gets way less out of his first serve. And since the majority of points in tennis are played following a first serve – you will rarely see a player with a first serve percentage below 50% – it follows that the low first serve win percentage is the driver behind Schwartzman holding serve less often than the other players in the top 10. The data bears that out – the closest player to Schwartzman’s 64.7% first serve win percentage is Rafael Nadal at 73.6% – about 9% higher.

Knowing that the key to holding serve more often will come from getting more out of the first serve, let’s look at Schwartzman’s first serve patters from 2020 and see if something stands out.

This is Schwartzman serving from the deuce side in 2020:

Below is the same data in a table. I have added a column “Free Point %,” which is just the number of aces and unreturned first serves divided by the total number of first serves hit into that particular part of the box.

Section	Deuce Wide	Deuce Mid	Deuce T
Serves Made	58.6%	12.5%	28.9%
Points Won	68.4%	50.8%	76.6%
Avg 1st Serve Speed	100 mph	101 mph	108 mph
Free Point %	26.2%	5%	31.8%

Data Courtesy of ATPTour.com

Looking at this table, the first adjustment I would try to make is to hit more first serves down the T in the deuce side of the court. It has been Schwartzman’s best first serve – hit at the highest velocity, with the highest overall winning percentage, and the highest free point percentage. Yet Schwartzman aimed that way less than 3 out of 10 times.

Let’s look at the same data for the ad side of the court:

Section	Ad Wide	Ad Mid	Ad T
Serves Made	31.2%	9.0%	59.8%
Points Won	65.9%	69.2%	59.5%
Avg 1st Serve Speed	103 mph	100 mph	104 mph
Free Point %	22.7%	21.9%	19.4%

Data Courtesy of ATPTour.com

In the ad side of the court, Schwartzman’s favorite serve was the T serve – to a right handed player’s forehand – and he hit that serve about 6 out of 10 times. Yet he had a higher winning percentage, and a higher free point percentage, going out wide. The difference in the ad side is smaller than it was in the deuce side, but it is something worth tracking and paying attention to.

Being one of the 10 best tennis players in the world is an impressive feat in and of itself. Yet if Schwartzman has his eyes set even higher, he will have to find a way to get more help from his first serve. A possible tweak – requiring no technical changes, or getting physically stronger – would be to aim more first serves down the T in the deuce, and out wide in the ad. Of course, it is not as simple as “prescribing” a player a certain serving pattern; the player has to be comfortable with the serve, and trust it in various match situations. But tennis is a game of adjustments, and this one might make holding serve a little easier for Mr. Schwartzman.

ATP Top 20: How Did They Get There?

As of December 31st, 2020, the average age in the ATP top 20 was 28.7. Roughly half of the players – 9 to be exact – have celebrated their 30th birthday; Roger Federer, Rafael Nadal, Stan Wawrinka, Gael Monfils, and others, have been mainstays in the second weeks of Grand Slams for over a decade. But how did these elite players get to the top of our sport in the first place? Specifically, how highly ranked were they as juniors? And after turning pro, how old were they when they first ranked inside the ATP top 100? How about the top 50, and finally top 20?

To do that, I have decided to look at the 20 highest ranked players in the final rankings for the years 2016 – 2020. During that period, 38 individual players have finished the year inside the top 20 in any given year. (Note that someone like, say, Felix Auger-Aliassime is not included in the dataset; while has has already been ranked inside the top 20, he hasn’t finished the year ranked as high just yet).

The raw data is presented below. While the column names are self-explanatory, I have decided to split the individual ages for a bit more granularity. For example, let’s say a particular player’s birthday is in May, and he cracks the top 100 in August of the year he turns 20. His value in the table would have been 20.25, indicating 20 years old “an a quarter.” This was done to differentiate players who achieve a certain milestone, let’s say, the day after their 19th birthday, as opposed to someone who achieves that same milestone at age 19 and 10 months. The players are listed in alphabetical order by last name.

2016 – 2020 EOY ATP Top 20

Name	Career High ITF Junior Ranking	ATP Top 100 Break Age	ATP Top 50 Break Age	ATP Top 20 Break Age
Anderson	28	22	24.75	27.25
Bautista Agut	47	24.25	25.25	26.25
Berdych	6	18.5	19	20.75
Berrettini	52	22	22.75	23.25
Carreno Busta	6	22	23.25	25.75
Cecchinato	100	22.75	25.75	26
Cilic	1	19	19.25	20.25
Coric	1	18	18.5	21.75
De Minaur	2	19.25	19.5	20.75
Del Potro	3	18	19	20
Dimitrov	1	19.75	21.5	22.75
Djokovic	24	18.25	19	19.25
Edmund	8	20.5	21.75	23.25
Federer	1	18	18.5	19.5
Fognini	8	20.5	24	26.25
Gasquet	1	17.25	19	19
Goffin	10	21.5	21.75	24.25
Isner	93	22.75	24.5	24.75
Karlovic	N/A	24.5	27.25	29
Khachanov	16	20	20.75	22.5
Kyrgios	1	19.25	19.75	21
Medvedev	13	20.75	21.5	22.75
Monfils	1	18.75	19	22
Murray	2	18.5	18.75	19.25
Nadal	145	16.75	17.25	18.75
Nishikori	7	18.25	21.25	22
Pouille	23	21.25	22.25	22.5
Querrey	10	19.25	19.75	22.75
Raonic	35	20	20.25	21.75
Rublev	1	19.75	20	22.25
Schwartzman	217	21.75	24.5	25.5
Shapovalov	2	18.25	18.5	20
Sock	22	20.75	22	24.25
Thiem	2	20.5	20.75	22
Tsitsipas	1	19.25	19.5	20
Tsonga	2	22.25	22.5	22.75
Wawrinka	7	20	20.75	23.25
Zverev	1	18	19	19.5

Mean		20.05	21.11	22.49
Median		19.875	20.75	22.375
Min		16.75	17.25	18.75
Max		24.5	27.25	29

There are three general takeaways from looking at the group as a whole:

High junior ranking
Out of the 38 players in the sample, 35 have been ranked inside the top 100 world junior rankings during the early stages of their tennis careers. The only exceptions are Ivo Karlovic, Diego Schwartzman, and Rafael Nadal. However, Rafael Nadal was already ranked inside the top 50 ATP by the time he turned 18 years old.

Moreover, 10 of the 38 players have been ranked #1 in the world as a junior, with 14 additional players having been ranked inside the top 10. That’s 24 out of the 38, or about 63% ranked higher that #10 ITF. A better than a top 100 junior ranking looks almost like a prerequisite to eventually make the ATP top 20, i.e these players have shown their potential in the junior ranks already.

Quick ascent into top 100
With most of these players having been elite juniors, they broke into the ATP top 100, on average, at around 20 years old. In other words, they tend to spend only about two years competing in the smaller tournaments on the Futures and Challenger circuits. Rafael Nadal was the youngest one in the sample, at less than 17 years old, while the oldest was Ivo Karlovic at over 24 years of age.

Slower ascent into top 20
As these players cracked the top 100 at around the age of 20, they first got into the top 50 roughly a year later, around the age of 21. The last thirty spots, into the top 20, then took about another year and a half, and the average age of cracking the top 20 was about 22.5. Once a player establishes himself in the top 20, he gets every opportunity – baring injury – to stay there, by being seeded at every tournament he plays and evading fellow top 20 players until the later rounds of a tournament. That is the main reason why the average age of the top 20 players in the world is significantly higher than the average age of the players breaking in for the first time: over the past five years at least, the players who make it into the top 20 tend to stay there.

While general observations are certainly useful, let’s see if we can learn some additional information by splitting the group of 38 players further. I have decided to do the split according to the highest achieved ITF junior ranking, and came up with the following three groups:

Players, who have achieved the #1 world ranking as juniors (10 players)
Players, who were ranked between #2 – #10 in the world as juniors (14 players)
Everyone else (14 players)

#1 ITF

Name	Career High ITF Junior Ranking	ATP Top 100 Break Age	ATP Top 50 Break Age	ATP Top 20 Break Age
Cilic	1	19	19.25	20.25
Coric	1	18	18.5	21.75
Dimitrov	1	19.75	21.5	22.75
Federer	1	18	18.5	19.5
Gasquet	1	17.25	19	19
Kyrgios	1	19.25	19.75	21
Monfils	1	18.75	19	22
Rublev	1	19.75	20	22.25
Tsitsipas	1	19.25	19.5	20
Zverev	1	18	19	19.5

Mean		18.7	19.4	20.8
Median		18.875	19.125	20.625
Min		17.25	18.5	19
Max		19.75	21.5	22.75

Not only has everyone on this list been ranked #1 ITF; every player on this list has won a junior Grand Slam title as well. In other words, these players would be everyone’s first choice in predicting the future stars of men’s tennis.

The #1 ranked juniors broke into the top 100 a year earlier than the whole group – 18.7 as opposed to 20.05. In fact, these guys tended to break into the top 50 by the time the rest of the group broke into the top 100. Interestingly enough, the pattern of a slower ascent into the top 20 holds for these players too: on average, it took them roughly 8 months (0.7 years) to make the jump from top 100 to top 50, but then another 17 months (1.4 years) to progress from top 50 to top 20.

If I was going to pick an upcoming member of this group, it would be the young Italian Lorenzo Musetti. Musetti has been ranked #1 ITF, he has won the 2019 Australian Open juniors crown, is currently ranked #129 ATP and doesn’t turn 19 until March 2021.

#2 – #10 ITF

Name	Career High ITF Junior Ranking	ATP Top 100 Break Age	ATP Top 50 Break Age	ATP Top 20 Break Age
Berdych	6	18.5	19	20.75
Carreno Busta	6	22	23.25	25.75
De Minaur	2	19.25	19.5	20.75
Del Potro	3	18	19	20
Edmund	8	20.5	21.75	23.25
Fognini	8	20.5	24	26.25
Goffin	10	21.5	21.75	24.25
Murray	2	18.5	18.75	19.25
Nishikori	7	18.25	21.25	22
Querrey	10	19.25	19.75	22.75
Shapovalov	2	18.25	18.5	20
Thiem	2	20.5	20.75	22
Tsonga	2	22.25	22.5	22.75
Wawrinka	7	20	20.75	23.25

Mean		19.80	20.75	22.36
Median		19.625	20.75	22.375
Min		18	18.5	19.25
Max		22.25	23.25	26.25

This group most closely follows the pattern of the whole sample, i.e. top 100 right around 20, top 50 around 21, and top 20 around 22.5 years old. On the other hand, there is a lot of variability within the group itself. We have players like Andy Murray, who have made it to the top 20 before their 20th birthday, while we also have someone like Fabio Fognini, who didn’t achieve that ranking until after he was 26 years old.

I want to point out two differences between this group, and the group of players who have achieved the #1 junior ranking. First, the #1 group, on average, made the jump into the ATP top 100 a full year younger than the #2 – #10 group. Also, while it took the #1 group less than a year to crack the top 50 after first making it to the top 100, it took the #2 – #10 group almost a full year. Admittedly, the sample size is not very big, yet the ascent up the professional ranks seemed to slow down a little bit for the #2 – #10 group.

My pick for the next member of this group is Thiago Seyboth Wild from Brazil. Thiago is currently ranked #117 ATP, was ranked as high as #8 in the world as a junior, and he will turn 21 in March of 2021.

Outside Top 10 ITF

Name	Career High ITF Junior Ranking	ATP Top 100 Break Age	ATP Top 50 Break Age	ATP Top 20 Break Age
Anderson	28	22	24.75	27.25
Bautista Agut	47	24.25	25.25	26.25
Berrettini	52	22	22.75	23.25
Cecchinato	100	22.75	25.75	26
Djokovic	24	18.25	19	19.25
Isner	93	22.75	24.5	24.75
Karlovic	N/A	24.5	27.25	29
Khachanov	16	20	20.75	22.5
Medvedev	13	20.75	21.5	22.75
Nadal	145	16.75	17.25	18.75
Pouille	23	21.25	22.25	22.5
Raonic	35	20	20.25	21.75
Schwartzman	217	21.75	24.5	25.5
Sock	22	20.75	22	24.25

Mean		21.27	22.70	23.84
Mean (no Nadal)		21.62	23.12	24.23

Median		21.5	22.5	23.75
Median (no Nadal)		21.75	22.75	24.25

Min		16.75	17.25	18.75
Min (no Nadal)		18.25	19	19.25

Max		24.5	27.25	29

This is the most varied group of the set. It includes players, who went to play NCAA Division 1 tennis after their junior careers, in John Isner and Kevin Anderson. It includes two of the best players of all time in Novak Djokovic and Rafael Nadal. It also includes Ivo Karlovic and Diego Schwartzman – really as polar opposites as you can get in terms of physical stature, as well as game style.

I have decided to calculate the statistics twice for this group: once with Rafael Nadal, and once without him. As he often is in tennis, Nadal is an extreme outlier, and could skew the data a little bit. I think that the statistics without Nadal are a little more informative.

Without Rafael Nadal, players in this group tended to break into the top 100, on average, after their 21st birthday, and into the top 20 around the age of 24. Compared to the #1 ranked juniors, their developments took roughly three years longer in both instances. Compared to the #2 – #10 group, their rankings progression was roughly two years slower.

I would guess that the main reason for this is – apart from the quality of the players – that players ranked outside of the top 10 ITF are less attractive for sponsors and agents, and are often not afforded the luxury of multiple wild cards into main draws of challengers or ATP tournaments. These guys have to earn their ranking “the hard way,” working their way through qualifying events at the Futures and Challenger levels. It takes a certain level of physical and mental maturity to be able to win multiple matches in a qualifying draw before even being afforded the opportunity to compete for ATP points in the main draw of a tournament. Junior and professional tennis are really two different worlds, and the adjustment to the pro game takes some time for the vast majority of players.

Takeaways

While I don’t think we can make sweeping generalizations based on this data alone, I think it contains some nuggets that we can keep in mind when looking for the next big star.

If you’re a young player, and your goal is to be in the top 20 in the world as a professional, a necessary first step is to achieve a top 100 junior ranking as a bare minimum; top 10 is really what you should be shooting for.

If you’re a tennis federation with a junior player ranked between #10 – #100 in the world, understand that he might not crack the top 100 for four or five years after his junior career is over. In other words, be prepared to support him for longer than one or two years, if you really believe in the player’s potential. Otherwise you run the risk of giving up too quickly.

Similarly, if you’re a parent of a player ranked between #10 – #100 ITF, and have the financial means to support your son for a year and then “see how it goes,” understand that your son is not likely to be in the top 100 after one year of professional tennis. It will be a longer process.

Finally, if you’re a fan and want to know the “next big thing” before the rest of your friends do, then pick a player, who has been recently ranked #1 in the world as a junior, and won a junior Grand Slam title. In about a year, your friends might be impressed.

Zverev and Rublev Should Pick Up the Pace

There is a saying in tennis that you are only as good as your second serve. We can see the validity of that statement across all levels of the sport. Go to your local park, and you’re likely to see aggressive swings on the first serve, followed by a tentative tap on the second serve. At the highest level of the sport, the second serve becomes the battleground where the all-valuable breaks of serve get decided.

One component of an effective second serve is its velocity. All else being equal, a faster second serve will be more effective: it gets to the opponent faster, giving him/her less time to react, and hit a good return. Using data from the 2020 Nitto ATP Finals, I’d like to highlight a possible area of improvement for two of the young starts of our game: Alexander Zverev and Andrey Rublev.

Alexander Zverev is still only 23 years old, yet he has already been ranked as high as #3 in the world, and came tantalizingly close to winning his first Grand Slam title at the 2020 US Open. Andrey Rublev is also just 23 years old, and recently broke into the top ten of the ATP rankings for the first time. Both of them have (hopefully) more than a decade of high level tennis ahead of them, and plenty of time to add to their game. I would argue that being more aggressive on their second serves could help them close the gap on their competitors.

First, let’s look at how aggressive – in terms of comparing their second serve velocities to their first serve velocities – the two of them were at the 2020 Nitto ATP Finals. In each of the tables below, the first two columns highlight the average first serve and second serve velocities in their matches in London. The third column is just the average second serve velocity subtracted from the average first serve velocity. And the fourth column is the difference in km/h divided by the average first serve speed.

Let’s look at Alexander Zverev’ table for an example. In his group stage match against Diego Schwartzman, he hit his first serve with an average speed of 215.3 km/h, while hitting his second serve with an average speed of 150.1 km/h. That means, on average, Zverev’s second serve was about 30% slower than his first serve in that particular match.

Alexander Zverev

Match	Average 1st Serve Speed (km/h)	Average 2nd Serve Speed (km/h)	Difference (km/h)	Difference (percentage)
Schwartzman (Group)	215.3	150.1	65.2	30.28%
Medvedev (Group)	209.0	151.3	57.7	27.61%
Djokovic (Group)	210.3	158.2	52.1	24.77%

Data courtesy of ATPTour.com

Andrey Rublev

Match	Average 1st Serve Speed (km/h)	Average 2nd Serve Speed (km/h)	Difference (km/h)	Difference (percentage)
Tsitsipas (Group)	197.0	144.4	52.6	26.70%
Thiem (Group)	198.4	147.1	51.3	25.86%
Nadal (Group)	192.4	150.5	41.9	21.78%

Data courtesy of ATPTour.com

Two things immediately stand out. First of all, both Zverev and Rublev generate outstanding velocities on their first serves. For the three matches he played in London, Zverev was sitting (!) at around 210 km/h with his first serve. Yet both of them hit their second serves roughly 50 km/h slower than their first serves, with the lone exception being Rublev’s match against Nadal. In that particular match, part of the reason for the smaller difference was Rublev’s decrease in his average first serve velocity by about 5 km/h. Without looking at the data, I would guess that Rublev hit more slice first serves to target the Nadal backhand return in both deuce and ad sides of the court; the slice serve tends to be hit slower than the typical flat first serve.

Knowing that Rublev and Zverev lost in the ballpark of 25% of their velocity on their second serves in the majority of their matches, let’s compare them to the four players, who have made it out of the group stage in London.

Dominic Thiem

Match	Average 1st Serve Speed (km/h)	Average 2nd Serve Speed (km/h)	Difference (km/h)	Difference (percentage)
Tsitsipas (Group)	192.3	163.7	28.6	14.87%
Rublev (Group)	193.6	166.9	26.7	13.79%
Nadal (Group)	193.5	163.6	29.9	15.45%
Djokovic (SF)	195.0	162.3	32.7	16.77%
Medvedev (F)	195.1	161.7	33.4	17.12%

Data courtesy of ATPTour.com

Rafael Nadal

Match	Average 1st Serve Speed (km/h)	Average 2nd Serve Speed (km/h)	Difference (km/h)	Difference (percentage)
Thiem (Group)	190.8	161.7	29.1	15.25%
Rublev (Group)	192.9	160.1	32.8	17.00%
Tsitsipas (Group)	192.1	162.4	29.7	15.46%
Medvedev (SF)	187.6	158.7	28.9	15.41%

Data courtesy of ATPTour.com

Neither Nadal, nor Thiem, hit their first serves as fast as Zverev or Rublev did. Specifically, Zverev hit his first serve in London on average 20 km/h faster than Thiem. Yet Thiem averaged over 160 km/h on his second serve in all of his matches, while Zverev didn’t crack 160 km/h once. In terms of the percentage differential, neither Nadal nor Thiem have lost more than 18% of their velocity on their second serves in any of their matches, while Rublev and Zverev have never lost less than 20%.

It could be the case that Nadal and Thiem have just been extremely aggressive on their second serves. So how do Rublev and Zverev compare to the other two semifinalists: Novak Djokovic and Daniil Medvedev?

Novak Djokovic

Match	Average 1st Serve Speed (km/h)	Average 2nd Serve Speed (km/h)	Difference (km/h)	Difference (percentage)
Medvedev (Group)	196.2	157.5	38.7	19.72%
Schwartzman (Group)	193.7	162.4	31.3	16.16%
Zverev (Group)	195.4	156.8	38.6	19.75%
Thiem (SF)	193.6	145.0	48.6	25.10%

Data courtesy of ATPTour.com

Daniil Medvedev

Match	Average 1st Serve Speed (km/h)	Average 2nd Serve Speed (km/h)	Difference (km/h)	Difference (percentage)
Schwartzman (Group)	199.1	160.8	38.3	19.24%
Zverev (Group)	196.4	157.4	39.0	19.86%
Djokovic (Group)	202.3	160.7	41.6	20.56%
Nadal (SF)	198.0	161.1	36.9	18.64%
Thiem (F)	200.5	158.3	42.2	21.05%

Data courtesy of ATPTour.com

Looking at their respective tables, Djokovic and Medvedev were more conservative with their second serves than Nadal and Thiem. Medvedev hit his second serve fairly consistently about 20% slower than his first serve. For Djokovic, there is more variability in his second serve velocities, most likely due to tactical reasons. The important takeaway here is that Djokovic and Medvedev lost 25% of their velocity on their second serve in just 1 of their 9 combined matches in London . Zverev and Rublev lost more than 25% on their second serve in 4 of the 6 combined matches they played.

Maybe Zverev and Rublev don’t have to be as aggressive with their second serves as Nadal and Thiem have been. By the same token, losing 25% of velocity on one’s second serve, as compared to some of the other elite players, leaves something on the table.

A good first step might be to close that gap to the roughly 20% we have seen from Djokovic and Medvedev in London. In Zverev’s case, assuming his first serve averages 210 km/h, that translates to a 168 km/h second serve. For Rublev, assuming a 195 km/h first serve, that would be a 156 km/h second serve. Is that attainable, or too ambitious? That’s for Rublev, Zverev, and their respective coaching staffs to decide. But I would sure like to see them try to be as aggressive on their second serves as they are with the rest of their games.